megre from upstream

.gitignore

@@ -1,6 +1,52 @@
-.build/
+# OSX leaves these everywhere on SMB shares
+._*
 
-# IntelliJ
+# OSX trash
-*.iml
+.DS_Store
+
+# Eclipse files
+.classpath
+.project
+.settings/**
+
+# Files generated by JetBrains IDEs, e.g. IntelliJ IDEA
 .idea/
+*.iml
 
+# Vscode files
+.vscode
+
+# Emacs save files
+*~
+\#*\#
+.\#*
+
+# Vim-related files
+[._]*.s[a-w][a-z]
+[._]s[a-w][a-z]
+*.un~
+Session.vim
+.netrwhist
+
+# Go test binaries
+*.test
+
+# Mercurial files
+**/.hg
+**/.hg*
+
+# Vagrant
+.vagrant
+network_closure.sh
+
+# make-related metadata
+/.make/
+
+# Terraform plans get put here
+/out/
+
+# Docker _src sync
+/docker/_src
+
+# build stuff
+.build

CHANGES.md

@@ -1,3 +1,20 @@
+## Aug 11 2016
+
+Reworked SSH keys and support for running CI builds
+
+* SSH keys are now stored as secrets.  `--ssh-public-key` will be created when you do `kops create cluster`.
+  You no longer need to specify a `--ssh-public-key` when you do an update, but if you do it will be imported.
+* An SSH public key must exist for AWS, if you do not have one you can import one with:
+  `kops create secret --name $CLUSTER_NAME sshpublickey admin -i ~/.ssh/id_rsa.pub`
+* For AWS, only a single SSH key can be used; you can delete extra keys with `kops delete secret`
+* To support changing SSH keys reliably, the name of the imported AWS SSH keypair will change to include
+  the OpenSSH key fingerprint.  Existing clusters will continue to work, but you will likely be prompted to
+  do a rolling update when you would otherwise not have to.  I suggest waiting till you next upgrade kubernetes.
+
+* Builds that are not published as Docker images can be run.  `kops` will pass a list of images in the NodeUp
+  configuration, and NodeUp will download and `docker load` these images.  For examples, see the
+  [testing tips](docs/testing.md)
+
 ## Jul 21 2016
 
 More rational model/UX - `kops create cluster` just creates spec, `kops update cluster` does real creation:

Makefile

@@ -1,22 +1,28 @@
-all: gocode
+all: kops
 
 DOCKER_REGISTRY=gcr.io/must-override/
 S3_BUCKET=s3://must-override/
 GOPATH_1ST=$(shell echo ${GOPATH} | cut -d : -f 1)
 
 ifndef VERSION
-  VERSION := git-$(shell git rev-parse --short HEAD)
+	VERSION := git-$(shell git rev-parse --short HEAD)
 endif
 
-gocode:
+kops:
-	GO15VENDOREXPERIMENT=1 go install -ldflags "-X main.BuildVersion=${VERSION}" k8s.io/kops/cmd/...
+	GO15VENDOREXPERIMENT=1 go install -ldflags "-X main.BuildVersion=${VERSION}" k8s.io/kops/cmd/kops/...
 	ln -sfn ${GOPATH_1ST}/src/k8s.io/kops/upup/models/ ${GOPATH_1ST}/bin/models
 
-# Build in a docker container with golang 1.5
+# Build in a docker container with golang 1.X
-# Used to test we have not broken 1.5
+# Used to test we have not broken 1.X
 check-builds-in-go15:
 	docker run -v ${GOPATH_1ST}/src/k8s.io/kops:/go/src/k8s.io/kops golang:1.5 make -f /go/src/k8s.io/kops/Makefile gocode
 
+check-builds-in-go16:
+	docker run -v ${GOPATH_1ST}/src/k8s.io/kops:/go/src/k8s.io/kops golang:1.6 make -f /go/src/k8s.io/kops/Makefile gocode
+
+check-builds-in-go17:
+	docker run -v ${GOPATH_1ST}/src/k8s.io/kops:/go/src/k8s.io/kops golang:1.7 make -f /go/src/k8s.io/kops/Makefile gocode
+
 codegen:
 	GO15VENDOREXPERIMENT=1 go install k8s.io/kops/upup/tools/generators/...
 	GO15VENDOREXPERIMENT=1 go generate k8s.io/kops/upup/pkg/fi/cloudup/awstasks
@@ -38,7 +44,7 @@ gofmt:
 	gofmt -w -s dns-controller/cmd
 	gofmt -w -s dns-controller/pkg
 
-kops-tar: gocode
+kops-tar: kops
 	rm -rf .build/kops/tar
 	mkdir -p .build/kops/tar/kops/
 	cp ${GOPATH_1ST}/bin/kops .build/kops/tar/kops/kops
@@ -47,15 +53,6 @@ kops-tar: gocode
 	tar tvf .build/kops.tar.gz
 	(sha1sum .build/kops.tar.gz | cut -d' ' -f1) > .build/kops.tar.gz.sha1
 
-nodeup-tar: gocode
-	rm -rf .build/nodeup/tar
-	mkdir -p .build/nodeup/tar/nodeup/root
-	cp ${GOPATH_1ST}/bin/nodeup .build/nodeup/tar/nodeup/root
-	cp -r upup/models/nodeup/ .build/nodeup/tar/nodeup/root/model/
-	tar czvf .build/nodeup.tar.gz -C .build/nodeup/tar/ .
-	tar tvf .build/nodeup.tar.gz
-	(sha1sum .build/nodeup.tar.gz | cut -d' ' -f1) > .build/nodeup.tar.gz.sha1 
-
 upload: nodeup-tar kops-tar
 	rm -rf .build/s3
 	mkdir -p .build/s3/nodeup
@@ -101,6 +98,27 @@ protokube-push: protokube-image
 
 
 
+nodeup: nodeup-tar
+
+nodeup-gocode:
+	go install -ldflags "-X main.BuildVersion=${VERSION}" k8s.io/kops/cmd/nodeup
+
+nodeup-builder-image:
+	docker build -f images/nodeup-builder/Dockerfile -t nodeup-builder .
+
+nodeup-build-in-docker: nodeup-builder-image
+	docker run -it -v `pwd`:/src nodeup-builder /onbuild.sh
+
+nodeup-tar: nodeup-build-in-docker
+	rm -rf .build/nodeup/tar
+	mkdir -p .build/nodeup/tar/nodeup/root
+	cp .build/artifacts/nodeup .build/nodeup/tar/nodeup/root
+	cp -r upup/models/nodeup/ .build/nodeup/tar/nodeup/root/model/
+	tar czvf .build/nodeup.tar.gz -C .build/nodeup/tar/ .
+	tar tvf .build/nodeup.tar.gz
+	(sha1sum .build/nodeup.tar.gz | cut -d' ' -f1) > .build/nodeup.tar.gz.sha1
+
+
 
 dns-controller-gocode:
 	go install k8s.io/kops/dns-controller/cmd/dns-controller
@@ -121,4 +139,3 @@ dns-controller-push: dns-controller-image
 
 copydeps:
 	rsync -avz _vendor/ vendor/ --exclude vendor/  --exclude .git
-

README.md

@@ -15,6 +15,10 @@ Some of the more interesting features:
 
 ## Recent changes
 
+Improvements are being made almost daily, but bigger changes are described here (until we get to
+a more formal release process):
+
+* Reworked SSH keys and support for running CI builds [Aug 11 2016](CHANGES.md#aug-11-2016)
 * Create command was split into create and update [Jul 21 2016](CHANGES.md#jul-21-2016)
 
 ## Installation
@@ -133,38 +137,11 @@ Each file in the tree describes a Task.
 On the nodeup side, Tasks can manage files, systemd services, packages etc.
 On the `kops update cluster` side, Tasks manage cloud resources: instances, networks, disks etc.
 
-## Workaround for terraform bug
+## Generate a terraform configuration
 
-Terraform currently has a bug where it can't create AWS tags containing a dot.  Until this is fixed,
+Kops can also generate a terraform configuration, which you can then apply using terraform, to build a Kubernetes
-you can't use terraform to build EC2 resources that are tagged with `k8s.io/...` tags.  Thankfully this is only
+cluster using terraform.
-the volumes, and it isn't the worst idea to build these separately anyway.
 
-We divide the cloudup model into three parts:
+If you are using a version of terraform prior to 0.7, please read about the [workaround for earlier versions of terraform](docs/terraform.md).
-* models/config which contains all the options - this is run automatically by "create cluster"
-* models/proto which sets up the volumes and other data which would be hard to recover (e.g. likely keys & secrets in the near future)
-* models/cloudup which is the main cloud model for configuring everything else
 
-So you don't use terraform for the 'proto' phase (you can't anyway, because of the bug!):
+For more details, please read the [how to use terraform to create a Kubernetes cluster](docs/terraform.md)
-
-```
-export KOPS_STATE_STORE=s3://<somes3bucket>
-export NAME=<kubernetes.mydomain.com>
-${GOPATH}/bin/kops create cluster --v=0 --zones=us-east-1c ${NAME}
-${GOPATH}/bin/kops update cluster --v=0 ${NAME} --model=proto --yes
-```
-
-And then you can use terraform to do the remainder of the installation:
-
-```
-export CLUSTER_NAME=<kubernetes.mydomain.com>
-${GOPATH}/bin/kops update cluster --v=0 ${NAME} --model=cloudup --target=terraform
-```
-
-Then, to apply using terraform:
-
-```
-cd out/terraform
-
-terraform plan
-terraform apply
-```

cmd/kops/create_cluster.go

@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"github.com/golang/glog"
 	"github.com/spf13/cobra"
+	"io/ioutil"
 	"k8s.io/kops/upup/pkg/api"
 	"k8s.io/kops/upup/pkg/fi"
 	"k8s.io/kops/upup/pkg/fi/cloudup"
@@ -15,24 +16,25 @@ import (
 )
 
 type CreateClusterCmd struct {
-	Yes                 bool
+	Yes               bool
-	Target              string
+	Target            string
-	Models              string
+	Models            string
-	Cloud               string
+	Cloud             string
-	Zones               string
+	Zones             string
-	MasterZones         string
+	MasterZones       string
-	NodeSize            string
+	NodeSize          string
-	MasterSize          string
+	MasterSize        string
-	NodeCount           int
+	NodeCount         int
-	Project             string
+	Project           string
-	KubernetesVersion   string
+	KubernetesVersion string
-	OutDir              string
+	OutDir            string
-	Image               string
+	Image             string
-	SSHPublicKey        string
+	SSHPublicKey      string
-	VPCID               string
+	VPCID             string
-	NetworkCIDR         string
+	NetworkCIDR       string
-	DNSZone             string
+	DNSZone           string
-	AdminAccess         string
+	AdminAccess       string
+	Networking        string
 	NoAssociatePublicIP bool
 }
 
@@ -78,6 +80,8 @@ func init() {
 
 	cmd.Flags().StringVar(&createCluster.Image, "image", "", "Image to use")
 
+	cmd.Flags().StringVar(&createCluster.Networking, "networking", "classic", "Networking mode to use.  kubenet, classic, external.  This currently defaults to classic, but will likely default to kubenet soon.")
+
 	cmd.Flags().StringVar(&createCluster.DNSZone, "dns-zone", "", "DNS hosted zone to use (defaults to last two components of cluster name)")
 	cmd.Flags().StringVar(&createCluster.OutDir, "out", "", "Path to write any local output")
 	cmd.Flags().StringVar(&createCluster.AdminAccess, "admin-access", "", "Restrict access to admin endpoints (SSH, HTTPS) to this CIDR.  If not set, access will not be restricted by IP.")
@@ -93,15 +97,16 @@ func (c *CreateClusterCmd) Run(args []string) error {
 
 	isDryrun := false
 	// direct requires --yes (others do not, because they don't make changes)
+	targetName := c.Target
 	if c.Target == cloudup.TargetDirect {
 		if !c.Yes {
 			isDryrun = true
-			c.Target = cloudup.TargetDryRun
+			targetName = cloudup.TargetDryRun
 		}
 	}
 	if c.Target == cloudup.TargetDryRun {
 		isDryrun = true
-		c.Target = cloudup.TargetDryRun
+		targetName = cloudup.TargetDryRun
 	}
 
 	clusterName := rootCommand.clusterName
@@ -132,6 +137,17 @@ func (c *CreateClusterCmd) Run(args []string) error {
 	cluster = &api.Cluster{}
 	var instanceGroups []*api.InstanceGroup
 
+	cluster.Spec.Networking = &api.NetworkingSpec{}
+	switch c.Networking {
+	case "classic":
+		cluster.Spec.Networking.Classic = &api.ClassicNetworkingSpec{}
+	case "kubenet":
+		cluster.Spec.Networking.Kubenet = &api.KubenetNetworkingSpec{}
+	case "external":
+		cluster.Spec.Networking.External = &api.ExternalNetworkingSpec{}
+	default:
+		return fmt.Errorf("unknown networking mode %q", c.Networking)
+	}
 	if c.Zones != "" {
 		existingZones := make(map[string]*api.ClusterZoneSpec)
 		for _, zone := range cluster.Spec.Zones {
@@ -304,8 +320,14 @@ func (c *CreateClusterCmd) Run(args []string) error {
 		}
 	}
 
+	sshPublicKeys := make(map[string][]byte)
 	if c.SSHPublicKey != "" {
 		c.SSHPublicKey = utils.ExpandPath(c.SSHPublicKey)
+		authorized, err := ioutil.ReadFile(c.SSHPublicKey)
+		if err != nil {
+			return fmt.Errorf("error reading SSH key file %q: %v", c.SSHPublicKey, err)
+		}
+		sshPublicKeys[fi.SecretNameSSHPrimary] = authorized
 	}
 
 	if c.AdminAccess != "" {
@@ -357,6 +379,14 @@ func (c *CreateClusterCmd) Run(args []string) error {
 		return fmt.Errorf("error writing completed cluster spec: %v", err)
 	}
 
+	for k, data := range sshPublicKeys {
+		keyStore := clusterRegistry.KeyStore(cluster.Name)
+		err = keyStore.AddSSHPublicKey(k, data)
+		if err != nil {
+			return fmt.Errorf("error addding SSH public key: %v", err)
+		}
+	}
+
 	if isDryrun {
 		fmt.Println("Previewing changes that will be made:\n")
 	}
@@ -366,8 +396,7 @@ func (c *CreateClusterCmd) Run(args []string) error {
 		InstanceGroups:  fullInstanceGroups,
 		Models:          strings.Split(c.Models, ","),
 		ClusterRegistry: clusterRegistry,
-		Target:          c.Target,
+		TargetName:      targetName,
-		SSHPublicKey:    c.SSHPublicKey,
 		OutDir:          c.OutDir,
 		DryRun:          isDryrun,
 	}
@@ -399,6 +428,7 @@ func (c *CreateClusterCmd) Run(args []string) error {
 		x := &kutil.CreateKubecfg{
 			ClusterName:      cluster.Name,
 			KeyStore:         clusterRegistry.KeyStore(cluster.Name),
+			SecretStore:      clusterRegistry.SecretStore(cluster.Name),
 			MasterPublicName: cluster.Spec.MasterPublicName,
 		}
 		defer x.Close()

cmd/kops/delete_cluster.go

@@ -56,7 +56,7 @@ func (c *DeleteClusterCmd) Run(args []string) error {
 
 	clusterName := rootCommand.clusterName
 	if clusterName == "" {
-		return fmt.Errorf("--name is required (when --external)")
+		return fmt.Errorf("--name is required (for safety)")
 	}
 
 	var cloud fi.Cloud

cmd/kops/delete_secret.go

@@ -57,7 +57,7 @@ func (c *DeleteSecretCmd) Run(args []string) error {
 	}
 
 	if len(secrets) == 0 {
-		return fmt.Errorf("secret %q not found")
+		return fmt.Errorf("secret not found")
 	}
 
 	if len(secrets) != 1 {

cmd/kops/edit_cluster.go

@@ -45,6 +45,11 @@ func (c *EditClusterCmd) Run(args []string) error {
 		return err
 	}
 
+	err = oldCluster.FillDefaults()
+	if err != nil {
+		return err
+	}
+
 	instanceGroupRegistry, err := rootCommand.InstanceGroupRegistry()
 	if err != nil {
 		return err

cmd/kops/export_kubecfg.go

@@ -54,6 +54,7 @@ func (c *ExportKubecfgCommand) Run(args []string) error {
 	x := &kutil.CreateKubecfg{
 		ClusterName:      clusterName,
 		KeyStore:         clusterRegistry.KeyStore(clusterName),
+		SecretStore:      clusterRegistry.SecretStore(cluster.Name),
 		MasterPublicName: master,
 	}
 	defer x.Close()

cmd/kops/get_cluster.go

@@ -4,7 +4,6 @@ import (
 	"os"
 
 	"fmt"
-	"github.com/golang/glog"
 	"github.com/spf13/cobra"
 	"k8s.io/kops/upup/pkg/api"
 	"strings"
@@ -25,7 +24,7 @@ func init() {
 		Run: func(cmd *cobra.Command, args []string) {
 			err := getClustersCmd.Run(args)
 			if err != nil {
-				glog.Exitf("%v", err)
+				exitWithError(err)
 			}
 		},
 	}

cmd/kops/kubectl.go

@@ -1,16 +0,0 @@
-package main
-
-type kubectlConfig struct {
-	Kind       string                    `json:"kind`
-	ApiVersion string                    `json:"apiVersion`
-	Clusters   []*kubectlClusterWithName `json:"clusters`
-}
-
-type kubectlClusterWithName struct {
-	Name    string         `json:"name`
-	Cluster kubectlCluster `json:"cluster`
-}
-
-type kubectlCluster struct {
-	Server string `json:"server`
-}

cmd/kops/rollingupdate_cluster.go

@@ -5,14 +5,14 @@ import (
 	"os"
 	"strconv"
 
-	"github.com/golang/glog"
 	"github.com/spf13/cobra"
 	"k8s.io/kops/upup/pkg/fi/cloudup"
 	"k8s.io/kops/upup/pkg/kutil"
 )
 
 type RollingUpdateClusterCmd struct {
-	Yes bool
+	Yes   bool
+	Force bool
 
 	cobraCommand *cobra.Command
 }
@@ -29,12 +29,13 @@ func init() {
 	cmd := rollingupdateCluster.cobraCommand
 	rollingUpdateCommand.cobraCommand.AddCommand(cmd)
 
-	cmd.Flags().BoolVar(&rollingupdateCluster.Yes, "yes", false, "Rollingupdate without confirmation")
+	cmd.Flags().BoolVar(&rollingupdateCluster.Yes, "yes", false, "perform rolling update without confirmation")
+	cmd.Flags().BoolVar(&rollingupdateCluster.Force, "force", false, "Force rolling update, even if no changes")
 
 	cmd.Run = func(cmd *cobra.Command, args []string) {
 		err := rollingupdateCluster.Run()
 		if err != nil {
-			glog.Exitf("%v", err)
+			exitWithError(err)
 		}
 	}
 }
@@ -107,8 +108,7 @@ func (c *RollingUpdateClusterCmd) Run() error {
 		}
 	}
 
-	if !needUpdate {
+	if !needUpdate && !c.Force {
-		// TODO: Allow --force option to force even if not needed?
 		fmt.Printf("\nNo rolling-update required\n")
 		return nil
 	}
@@ -118,5 +118,5 @@ func (c *RollingUpdateClusterCmd) Run() error {
 		return nil
 	}
 
-	return d.RollingUpdate(groups)
+	return d.RollingUpdate(groups, c.Force)
 }

cmd/kops/root.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"os"
 
-	"encoding/json"
 	"github.com/golang/glog"
 	"github.com/spf13/cobra"
 	"github.com/spf13/viper"
@@ -114,7 +113,7 @@ func (c *RootCmd) ClusterName() string {
 	return c.clusterName
 }
 
-func readKubectlClusterConfig() (*kubectlClusterWithName, error) {
+func readKubectlClusterConfig() (*kutil.KubectlClusterWithName, error) {
 	kubectl := &kutil.Kubectl{}
 	context, err := kubectl.GetCurrentContext()
 	if err != nil {
@@ -122,18 +121,12 @@ func readKubectlClusterConfig() (*kubectlClusterWithName, error) {
 	}
 	glog.V(4).Infof("context = %q", context)
 
-	configString, err := kubectl.GetConfig(true, "json")
+	config, err := kubectl.GetConfig(true)
 	if err != nil {
 		return nil, fmt.Errorf("error getting current config from kubectl: %v", err)
 	}
-	glog.V(8).Infof("config = %q", configString)
-
-	config := &kubectlConfig{}
-	err = json.Unmarshal([]byte(configString), config)
-	if err != nil {
-		return nil, fmt.Errorf("cannot parse current config from kubectl: %v", err)
-	}
 
+	// Minify should have done this
 	if len(config.Clusters) != 1 {
 		return nil, fmt.Errorf("expected exactly one cluster in kubectl config, found %d", len(config.Clusters))
 	}
@@ -155,6 +148,10 @@ func (c *RootCmd) ClusterRegistry() (*api.ClusterRegistry, error) {
 		return nil, fmt.Errorf("error building state store path for %q: %v", c.stateLocation, err)
 	}
 
+	if !vfs.IsClusterReadable(basePath) {
+		return nil, fmt.Errorf("State store %q is not cloud-reachable - please use an S3 bucket", c.stateLocation)
+	}
+
 	clusterRegistry := api.NewClusterRegistry(basePath)
 	c.clusterRegistry = clusterRegistry
 	return clusterRegistry, nil
@@ -186,7 +183,7 @@ func (c *RootCmd) Cluster() (*api.ClusterRegistry, *api.Cluster, error) {
 }
 
 func (c *RootCmd) InstanceGroupRegistry() (*api.InstanceGroupRegistry, error) {
-	clusterStore, err := c.ClusterRegistry()
+	clusterRegistry, err := c.ClusterRegistry()
 	if err != nil {
 		return nil, err
 	}
@@ -196,11 +193,11 @@ func (c *RootCmd) InstanceGroupRegistry() (*api.InstanceGroupRegistry, error) {
 		return nil, fmt.Errorf("--name is required")
 	}
 
-	return clusterStore.InstanceGroups(clusterName)
+	return clusterRegistry.InstanceGroups(clusterName)
 }
 
 func (c *RootCmd) SecretStore() (fi.SecretStore, error) {
-	clusterStore, err := c.ClusterRegistry()
+	clusterRegistry, err := c.ClusterRegistry()
 	if err != nil {
 		return nil, err
 	}
@@ -210,11 +207,11 @@ func (c *RootCmd) SecretStore() (fi.SecretStore, error) {
 		return nil, fmt.Errorf("--name is required")
 	}
 
-	return clusterStore.SecretStore(clusterName), nil
+	return clusterRegistry.SecretStore(clusterName), nil
 }
 
 func (c *RootCmd) KeyStore() (fi.CAStore, error) {
-	clusterStore, err := c.ClusterRegistry()
+	clusterRegistry, err := c.ClusterRegistry()
 	if err != nil {
 		return nil, err
 	}
@@ -224,5 +221,5 @@ func (c *RootCmd) KeyStore() (fi.CAStore, error) {
 		return nil, fmt.Errorf("--name is required")
 	}
 
-	return clusterStore.KeyStore(clusterName), nil
+	return clusterRegistry.KeyStore(clusterName), nil
 }

cmd/kops/toolbox.go

@@ -0,0 +1,15 @@
+package main
+
+import (
+	"github.com/spf13/cobra"
+)
+
+// toolboxCmd represents the toolbox command
+var toolboxCmd = &cobra.Command{
+	Use:   "toolbox",
+	Short: "Misc infrequently used commands",
+}
+
+func init() {
+	rootCommand.AddCommand(toolboxCmd)
+}

cmd/kops/toolbox_convert_imported.go

@@ -0,0 +1,100 @@
+package main
+
+import (
+	"fmt"
+
+	"github.com/spf13/cobra"
+	"k8s.io/kops/upup/pkg/api"
+	"k8s.io/kops/upup/pkg/fi/cloudup/awsup"
+	"k8s.io/kops/upup/pkg/kutil"
+)
+
+type ConvertImportedCmd struct {
+	NewClusterName string
+}
+
+var convertImported ConvertImportedCmd
+
+func init() {
+	cmd := &cobra.Command{
+		Use:   "convert-imported",
+		Short: "Convert an imported cluster into a kops cluster",
+		Run: func(cmd *cobra.Command, args []string) {
+			err := convertImported.Run()
+			if err != nil {
+				exitWithError(err)
+			}
+		},
+	}
+
+	toolboxCmd.AddCommand(cmd)
+
+	cmd.Flags().StringVar(&convertImported.NewClusterName, "newname", "", "new cluster name")
+}
+
+func (c *ConvertImportedCmd) Run() error {
+	clusterRegistry, cluster, err := rootCommand.Cluster()
+	if err != nil {
+		return err
+	}
+
+	instanceGroupRegistry, err := rootCommand.InstanceGroupRegistry()
+	if err != nil {
+		return err
+	}
+
+	instanceGroups, err := instanceGroupRegistry.ReadAll()
+
+	if cluster.Annotations[api.AnnotationNameManagement] != api.AnnotationValueManagementImported {
+		return fmt.Errorf("cluster %q does not appear to be a cluster imported using kops import", cluster.Name)
+	}
+
+	if c.NewClusterName == "" {
+		return fmt.Errorf("--newname is required for converting an imported cluster")
+	}
+
+	oldClusterName := cluster.Name
+	if oldClusterName == "" {
+		return fmt.Errorf("(Old) ClusterName must be set in configuration")
+	}
+
+	// TODO: Switch to cloudup.BuildCloud
+	if len(cluster.Spec.Zones) == 0 {
+		return fmt.Errorf("Configuration must include Zones")
+	}
+
+	region := ""
+	for _, zone := range cluster.Spec.Zones {
+		if len(zone.Name) <= 2 {
+			return fmt.Errorf("Invalid AWS zone: %q", zone.Name)
+		}
+
+		zoneRegion := zone.Name[:len(zone.Name)-1]
+		if region != "" && zoneRegion != region {
+			return fmt.Errorf("Clusters cannot span multiple regions")
+		}
+
+		region = zoneRegion
+	}
+
+	tags := map[string]string{"KubernetesCluster": oldClusterName}
+	cloud, err := awsup.NewAWSCloud(region, tags)
+	if err != nil {
+		return fmt.Errorf("error initializing AWS client: %v", err)
+	}
+
+	d := &kutil.ConvertKubeupCluster{}
+	d.NewClusterName = c.NewClusterName
+	d.OldClusterName = oldClusterName
+	d.Cloud = cloud
+	d.ClusterConfig = cluster
+	d.InstanceGroups = instanceGroups
+	d.ClusterRegistry = clusterRegistry
+
+	err = d.Upgrade()
+	if err != nil {
+		return err
+	}
+
+	return nil
+}

cmd/kops/update_cluster.go

@@ -4,10 +4,13 @@ import (
 	"fmt"
 	"github.com/golang/glog"
 	"github.com/spf13/cobra"
+	"io/ioutil"
 	"k8s.io/kops/upup/pkg/api"
+	"k8s.io/kops/upup/pkg/fi"
 	"k8s.io/kops/upup/pkg/fi/cloudup"
 	"k8s.io/kops/upup/pkg/fi/utils"
 	"k8s.io/kops/upup/pkg/kutil"
+	"os"
 	"strings"
 )
 
@@ -39,7 +42,7 @@ func init() {
 	cmd.Flags().BoolVar(&updateCluster.Yes, "yes", false, "Actually create cloud resources")
 	cmd.Flags().StringVar(&updateCluster.Target, "target", "direct", "Target - direct, terraform")
 	cmd.Flags().StringVar(&updateCluster.Models, "model", "config,proto,cloudup", "Models to apply (separate multiple models with commas)")
-	cmd.Flags().StringVar(&updateCluster.SSHPublicKey, "ssh-public-key", "~/.ssh/id_rsa.pub", "SSH public key to use")
+	cmd.Flags().StringVar(&updateCluster.SSHPublicKey, "ssh-public-key", "", "SSH public key to use (deprecated: use kops create secret instead)")
 	cmd.Flags().StringVar(&updateCluster.OutDir, "out", "", "Path to write any local output")
 }
 
@@ -50,16 +53,18 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 	}
 
 	isDryrun := false
+	targetName := c.Target
+
 	// direct requires --yes (others do not, because they don't do anything!)
 	if c.Target == cloudup.TargetDirect {
 		if !c.Yes {
 			isDryrun = true
-			c.Target = cloudup.TargetDryRun
+			targetName = cloudup.TargetDryRun
 		}
 	}
 	if c.Target == cloudup.TargetDryRun {
 		isDryrun = true
-		c.Target = cloudup.TargetDryRun
+		targetName = cloudup.TargetDryRun
 	}
 
 	if c.OutDir == "" {
@@ -87,7 +92,21 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 	}
 
 	if c.SSHPublicKey != "" {
+		fmt.Fprintf(os.Stderr, "--ssh-public-key on update is deprecated - please use `kops create secret --name %s sshpublickey admin -i ~/.ssh/id_rsa.pub` instead\n", cluster.Name)
+
 		c.SSHPublicKey = utils.ExpandPath(c.SSHPublicKey)
+		authorized, err := ioutil.ReadFile(c.SSHPublicKey)
+		if err != nil {
+			return fmt.Errorf("error reading SSH key file %q: %v", c.SSHPublicKey, err)
+		}
+		keyStore, err := rootCommand.KeyStore()
+		if err != nil {
+			return err
+		}
+		err = keyStore.AddSSHPublicKey(fi.SecretNameSSHPrimary, authorized)
+		if err != nil {
+			return fmt.Errorf("error addding SSH public key: %v", err)
+		}
 	}
 
 	strict := false
@@ -101,8 +120,7 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 		InstanceGroups:  fullInstanceGroups,
 		Models:          strings.Split(c.Models, ","),
 		ClusterRegistry: clusterRegistry,
-		Target:          c.Target,
+		TargetName:      targetName,
-		SSHPublicKey:    c.SSHPublicKey,
 		OutDir:          c.OutDir,
 		DryRun:          isDryrun,
 	}
@@ -112,12 +130,23 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 	}
 
 	if isDryrun {
-		fmt.Printf("Must specify --yes to apply changes\n")
+		target := applyCmd.Target.(*fi.DryRunTarget)
+		if target.HasChanges() {
+			fmt.Printf("Must specify --yes to apply changes\n")
+		} else {
+			fmt.Printf("No changes need to be applied\n")
+		}
 		return nil
 	}
 
 	// TODO: Only if not yet set?
 	if !isDryrun {
+		hasKubecfg, err := hasKubecfg(cluster.Name)
+		if err != nil {
+			glog.Warningf("error reading kubecfg: %v", err)
+			hasKubecfg = true
+		}
+
 		keyStore := clusterRegistry.KeyStore(cluster.Name)
 
 		kubecfgCert, err := keyStore.FindCert("kubecfg")
@@ -131,6 +160,7 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 			x := &kutil.CreateKubecfg{
 				ClusterName:      cluster.Name,
 				KeyStore:         keyStore,
+				SecretStore:      clusterRegistry.SecretStore(cluster.Name),
 				MasterPublicName: cluster.Spec.MasterPublicName,
 			}
 			defer x.Close()
@@ -142,7 +172,35 @@ func (c *UpdateClusterCmd) Run(args []string) error {
 		} else {
 			glog.Infof("kubecfg cert not found; won't export kubecfg")
 		}
+
+		if !hasKubecfg {
+			// Assume initial creation
+			fmt.Printf("\n")
+			fmt.Printf("Cluster is starting.  It should be ready in a few minutes.\n")
+			fmt.Printf("\n")
+			fmt.Printf("Suggestions:\n")
+			fmt.Printf(" * list nodes: kubectl get nodes --show-labels\n")
+			fmt.Printf(" * ssh to the master: ssh -i ~/.ssh/id_rsa admin@%s\n", cluster.Spec.MasterPublicName)
+			fmt.Printf(" * read about installing addons: https://github.com/kubernetes/kops/blob/master/docs/addons.md\n")
+			fmt.Printf("\n")
+		}
 	}
 
 	return nil
 }
+
+func hasKubecfg(contextName string) (bool, error) {
+	kubectl := &kutil.Kubectl{}
+
+	config, err := kubectl.GetConfig(false)
+	if err != nil {
+		return false, fmt.Errorf("error getting config from kubectl: %v", err)
+	}
+
+	for _, context := range config.Contexts {
+		if context.Name == contextName {
+			return true, nil
+		}
+	}
+	return false, nil
+}

cmd/kops/upgrade_cluster.go

@@ -3,14 +3,14 @@ package main
 import (
 	"fmt"
 
-	"github.com/golang/glog"
 	"github.com/spf13/cobra"
-	"k8s.io/kops/upup/pkg/fi/cloudup/awsup"
+	"k8s.io/kops/upup/pkg/api"
-	"k8s.io/kops/upup/pkg/kutil"
+	"k8s.io/kops/upup/pkg/fi/cloudup"
+	"os"
 )
 
 type UpgradeClusterCmd struct {
-	NewClusterName string
+	Yes bool
 }
 
 var upgradeCluster UpgradeClusterCmd
@@ -23,21 +23,26 @@ func init() {
 		Run: func(cmd *cobra.Command, args []string) {
 			err := upgradeCluster.Run()
 			if err != nil {
-				glog.Exitf("%v", err)
+				exitWithError(err)
 			}
 		},
 	}
 
-	upgradeCmd.AddCommand(cmd)
+	cmd.Flags().BoolVar(&upgradeCluster.Yes, "yes", false, "Apply update")
 
-	cmd.Flags().StringVar(&upgradeCluster.NewClusterName, "newname", "", "new cluster name")
+	upgradeCmd.AddCommand(cmd)
+}
+
+type upgradeAction struct {
+	Item     string
+	Property string
+	Old      string
+	New      string
+
+	apply func()
 }
 
 func (c *UpgradeClusterCmd) Run() error {
-	if c.NewClusterName == "" {
-		return fmt.Errorf("--newname is required")
-	}
-
 	clusterRegistry, cluster, err := rootCommand.Cluster()
 	if err != nil {
 		return err
@@ -50,47 +55,96 @@ func (c *UpgradeClusterCmd) Run() error {
 
 	instanceGroups, err := instanceGroupRegistry.ReadAll()
 
-	oldClusterName := cluster.Name
+	if cluster.Annotations[api.AnnotationNameManagement] == api.AnnotationValueManagementImported {
-	if oldClusterName == "" {
+		return fmt.Errorf("upgrade is not for use with imported clusters (did you mean `kops toolbox convert-imported`?)")
-		return fmt.Errorf("(Old) ClusterName must be set in configuration")
 	}
 
-	if len(cluster.Spec.Zones) == 0 {
+	latestKubernetesVersion, err := api.FindLatestKubernetesVersion()
-		return fmt.Errorf("Configuration must include Zones")
-	}
-
-	region := ""
-	for _, zone := range cluster.Spec.Zones {
-		if len(zone.Name) <= 2 {
-			return fmt.Errorf("Invalid AWS zone: %q", zone.Name)
-		}
-
-		zoneRegion := zone.Name[:len(zone.Name)-1]
-		if region != "" && zoneRegion != region {
-			return fmt.Errorf("Clusters cannot span multiple regions")
-		}
-
-		region = zoneRegion
-	}
-
-	tags := map[string]string{"KubernetesCluster": oldClusterName}
-	cloud, err := awsup.NewAWSCloud(region, tags)
-	if err != nil {
-		return fmt.Errorf("error initializing AWS client: %v", err)
-	}
-
-	d := &kutil.UpgradeCluster{}
-	d.NewClusterName = c.NewClusterName
-	d.OldClusterName = oldClusterName
-	d.Cloud = cloud
-	d.ClusterConfig = cluster
-	d.InstanceGroups = instanceGroups
-	d.ClusterRegistry = clusterRegistry
-
-	err = d.Upgrade()
 	if err != nil {
 		return err
 	}
 
+	var actions []*upgradeAction
+	if cluster.Spec.KubernetesVersion != latestKubernetesVersion {
+		actions = append(actions, &upgradeAction{
+			Item:     "Cluster",
+			Property: "KubernetesVersion",
+			Old:      cluster.Spec.KubernetesVersion,
+			New:      latestKubernetesVersion,
+			apply: func() {
+				cluster.Spec.KubernetesVersion = latestKubernetesVersion
+			},
+		})
+	}
+
+	if len(actions) == 0 {
+		// TODO: Allow --force option to force even if not needed?
+		// Note stderr - we try not to print to stdout if no update is needed
+		fmt.Fprintf(os.Stderr, "\nNo upgrade required\n")
+		return nil
+	}
+
+	{
+		t := &Table{}
+		t.AddColumn("ITEM", func(a *upgradeAction) string {
+			return a.Item
+		})
+		t.AddColumn("PROPERTY", func(a *upgradeAction) string {
+			return a.Property
+		})
+		t.AddColumn("OLD", func(a *upgradeAction) string {
+			return a.Old
+		})
+		t.AddColumn("NEW", func(a *upgradeAction) string {
+			return a.New
+		})
+
+		err := t.Render(actions, os.Stdout, "ITEM", "PROPERTY", "OLD", "NEW")
+		if err != nil {
+			return err
+		}
+	}
+
+	if !c.Yes {
+		fmt.Printf("\nMust specify --yes to perform upgrade\n")
+		return nil
+	} else {
+		for _, action := range actions {
+			action.apply()
+		}
+
+		// TODO: DRY this chunk
+		err = cluster.PerformAssignments()
+		if err != nil {
+			return fmt.Errorf("error populating configuration: %v", err)
+		}
+
+		fullCluster, err := cloudup.PopulateClusterSpec(cluster, clusterRegistry)
+		if err != nil {
+			return err
+		}
+
+		err = api.DeepValidate(fullCluster, instanceGroups, true)
+		if err != nil {
+			return err
+		}
+
+		// Note we perform as much validation as we can, before writing a bad config
+		err = clusterRegistry.Update(cluster)
+		if err != nil {
+			return err
+		}
+
+		err = clusterRegistry.WriteCompletedConfig(fullCluster)
+		if err != nil {
+			return fmt.Errorf("error writing completed cluster spec: %v", err)
+		}
+
+		fmt.Printf("\nUpdates applied to configuration.\n")
+
+		// TODO: automate this step
+		fmt.Printf("You can now apply these changes, using `kops update cluster %s`\n", cluster.Name)
+	}
+
 	return nil
 }

cmd/nodeup/main.go

@@ -6,17 +6,29 @@ import (
 	"github.com/golang/glog"
 	"k8s.io/kops/upup/pkg/fi/nodeup"
 	"os"
+	"time"
 )
 
+var (
+	// value overwritten during build. This can be used to resolve issues.
+	BuildVersion = "0.1"
+)
+
+const retryInterval = 30 * time.Second
+
 func main() {
+	fmt.Printf("nodeup version %s\n", BuildVersion)
+
 	flagModel := "model"
 	flag.StringVar(&flagModel, "model", flagModel, "directory to use as model for desired configuration")
 	var flagConf string
 	flag.StringVar(&flagConf, "conf", "node.yaml", "configuration location")
-	var flagAssetDir string
+	var flagCacheDir string
-	flag.StringVar(&flagAssetDir, "assets", "/var/cache/nodeup", "the location for the local asset cache")
+	flag.StringVar(&flagCacheDir, "cache", "/var/cache/nodeup", "the location for the local asset cache")
 	var flagRootFS string
 	flag.StringVar(&flagRootFS, "rootfs", "/", "the location of the machine root (for running in a container)")
+	var flagRetries int
+	flag.IntVar(&flagRetries, "retries", -1, "maximum number of retries on failure: -1 means retry forever")
 
 	dryrun := false
 	flag.BoolVar(&dryrun, "dryrun", false, "Don't create cloud resources; just show what would be done")
@@ -34,17 +46,32 @@ func main() {
 		glog.Exitf("--conf is required")
 	}
 
-	cmd := &nodeup.NodeUpCommand{
+	retries := flagRetries
-		ConfigLocation: flagConf,
+
-		ModelDir:       flagModel,
+	for {
-		Target:         target,
+		cmd := &nodeup.NodeUpCommand{
-		AssetDir:       flagAssetDir,
+			ConfigLocation: flagConf,
-		FSRoot:         flagRootFS,
+			ModelDir:       flagModel,
+			Target:         target,
+			CacheDir:       flagCacheDir,
+			FSRoot:         flagRootFS,
+		}
+		err := cmd.Run(os.Stdout)
+		if err == nil {
+			fmt.Printf("success")
+			os.Exit(0)
+		}
+
+		if retries == 0 {
+			glog.Exitf("error running nodeup: %v", err)
+			os.Exit(1)
+		}
+
+		if retries > 0 {
+			retries--
+		}
+
+		glog.Warningf("got error running nodeup (will retry in %s): %v", retryInterval, err)
+		time.Sleep(retryInterval)
 	}
-	err := cmd.Run(os.Stdout)
-	if err != nil {
-		glog.Exitf("error running nodeup: %v", err)
-		os.Exit(1)
-	}
-	fmt.Printf("success")
 }

docker/Dockerfile

@@ -0,0 +1,26 @@
+FROM golang:1.6-alpine
+
+ARG KOPS_SRC=_src
+COPY $KOPS_SRC/ /go/src/k8s.io/kops/
+
+# KUBECTL_SOURCE: Change to kubernetes-dev/ci for CI
+ARG KUBECTL_SOURCE=kubernetes-release/release
+
+# KUBECTL_TRACK: Currently latest from KUBECTL_SOURCE. Change to latest-1.3.txt, etc. if desired.
+ARG KUBECTL_TRACK=latest.txt
+
+ARG KUBECTL_ARCH=linux/amd64
+
+RUN apk add --no-cache --update build-base curl git mercurial  --virtual .kops-deps && \
+    cd "${GOPATH}/src/k8s.io/kops" && make && \
+    MODELS=$(readlink "${GOPATH}/bin/models") && mv "${MODELS}" /usr/local/bin && rm "${GOPATH}/bin/models" && mv ${GOPATH}/bin/* /usr/local/bin && \
+    GITISH=$(git describe --always) && \
+    KUBECTL_VERSION=${KUBECTL_VERSION:-$(curl -SsL --retry 5 "https://storage.googleapis.com/${KUBECTL_SOURCE}/${KUBECTL_TRACK}")} && \
+    echo "=== Fetching kubectl ${KUBECTL_VERSION} ===" && \
+    curl -SsL --retry 5 "https://storage.googleapis.com/${KUBECTL_SOURCE}/${KUBECTL_VERSION}/${KUBECTL_ARCH}/kubectl" > /usr/local/bin/kubectl && \
+    chmod +x /usr/local/bin/kubectl && \
+    /usr/local/bin/kubectl version --client && \
+    cd / && rm -rf "${GOPATH}" && rm -rf /usr/local/go && rm /usr/local/bin/go-wrapper && apk del .kops-deps && \
+    echo "=== Built kops at ${GITISH}, fetched kubectl ${KUBECTL_VERSION} ==="
+
+CMD "/go/bin/kops"

docker/README.md

@@ -0,0 +1,15 @@
+# Running Kops in Docker
+
+The Dockerfile here is offered primarily as a way to build continuous
+integration versions of `kops` until we figure out how we want to
+release/package it.
+
+To use it, e.g. (assumes your `$HOME` is correct and that `$KOPS_STATE_STORE` is correct):
+```shell
+$ docker build -t kops .
+$ KOPS="docker run -v $HOME/.aws:/root/.aws:ro -v $HOME/.ssh:/root/.ssh:ro -v $HOME/.kube:/root/.kube -it kops kops --state=$KOPS_STATE_STORE"
+```
+
+This creates a shell variable that runs the `kops` container with `~/.aws` mounted in (for AWS credentials), `~/.ssh` mounted in (for SSH keys, for AWS specifically), and `~/.kube` mounted in (so `kubectl` can add newly created clusters).
+
+After this, you can just use `$KOPS` where you would generally use `kops`, e.g. `$KOPS get cluster`.

docker/build-squash-push.sh

@@ -0,0 +1,77 @@
+#!/bin/bash
+
+set -o errexit
+set -o nounset
+set -o pipefail
+
+readonly DOCKER_ROOT=$(dirname "${BASH_SOURCE}")
+readonly GITISH="$(git describe --always)"
+readonly ARCH=${BUILD_ARCH:-"linux/amd64"}
+readonly NAME=${BUILD_NAME:-"ci-${GITISH}-${ARCH/\//-}"} # e.g. ci-bef7faf-linux-amd64
+readonly TMPNAME="${NAME}-$(date +%s)" # e.g. ci-bef7faf-linux-amd64-12345678
+readonly TAG=${BUILD_DOCKER_TAG:-"b.gcr.io/kops-ci/kops:${NAME}"}
+readonly PUSH_TAG=${BUILD_PUSH_TAG:-"no"}
+readonly CLEAN_TAG=${BUILD_CLEAN_TAG:-"yes"}
+readonly TMPTAG="${TAG}-$(date +%s)"
+readonly LINK=${BUILD_LINK:-} # Also pushes to e.g. ci-{BUILD_LINK}-linux-amd64, i.e. for "latest"
+readonly SYMBOLIC_TAG=${BUILD_SYMBOLIC_TAG:-"b.gcr.io/kops-ci/kops:ci-${LINK}-${ARCH/\//-}"}
+
+if [[ "${ARCH}" != "linux/amd64" ]]; then
+  echo "!!! Alternate architecture build not supported yet. !!!"
+  exit 1
+fi
+
+if [[ -z "${GITISH}" ]]; then
+  echo "!!! git hash not found, are you sure you're in a git tree and git is installed? !!!"
+  git config -l
+  exit 1
+fi
+
+echo
+echo "=== Copying src to docker/_src ==="
+echo
+
+rsync -a --exclude=/docker/ "${DOCKER_ROOT}/.." "${DOCKER_ROOT}/_src"
+
+echo
+echo "=== Building at ${GITISH} for ${ARCH} ==="
+echo
+
+# Build -> $TMPTAG
+docker build -t "${TMPTAG}" --build-arg "KUBECTL_ARCH=${ARCH}" --force-rm=true --rm=true --pull=true --no-cache=true "${DOCKER_ROOT}"
+
+# Squash -> $TAG
+docker create --name="${TMPNAME}" "${TMPTAG}"
+docker export "${TMPNAME}" | docker import - "${TAG}"
+
+if [[ "${PUSH_TAG}" == "yes" ]]; then
+  echo
+  echo "=== Pushing ${TAG} ==="
+  echo
+
+  gcloud docker push "${TAG}"
+fi
+
+if [[ -n "${LINK}" ]]; then
+  echo
+  echo "=== Pushing ${SYMBOLIC_TAG} ==="
+  echo
+  docker tag "${TAG}" "${SYMBOLIC_TAG}"
+  gcloud docker push "${SYMBOLIC_TAG}"
+fi
+
+echo
+echo "=== Cleaning up ==="
+echo
+docker rm "${TMPNAME}" || true
+docker rmi -f "${TMPTAG}" || true
+if [[ -n "${LINK}" ]]; then
+  docker rmi -f "${SYMBOLIC_TAG}" || true
+fi
+if [[ "${CLEAN_TAG}" == "yes" ]]; then
+  docker rmi -f "${TAG}" || true
+else
+  echo
+  echo "=== ${TAG} leaked (BUILD_CLEAN_TAG not set) ==="
+  echo
+fi

docs/addons.md

@@ -17,14 +17,14 @@ Install using:
 kubectl create -f https://raw.githubusercontent.com/kubernetes/kops/master/addons/dashboard/v1.1.0.yaml
 ```
 
-And then navigate to `https://<clustername>/ui`
+And then navigate to `https://api.<clustername>/ui`
 
 (`/ui` is an alias to `https://<clustername>/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard`)
 
 The login credentials are:
 
 * Username: `admin`
-* Password: get by running `kops get secrets kube --type secret -oplaintext`
+* Password: get by running `kops get secrets kube --type secret -oplaintext` or `kubectl config view --minify`
 
 
 ### Monitoring - Standalone

docs/networking.md

@@ -11,4 +11,13 @@ Because k8s modifies the AWS routing table, this means that realistically kubern
 routing table, and thus it requires its own subnet.  It is theoretically possible to share a routing table
 with other infrastructure (but not a second cluster!), but this is not really recommended.
 
-kops will support other networking options as they add support for the daemonset method of deployment.
+kops will support other networking options as they add support for the daemonset method of deployment.
+
+
+kops currently supports 3 networking modes:
+
+* `classic` kubernetes native networking, done in-process
+* `kubenet` kubernetes native networking via a CNI plugin.  Also has less reliance on Docker's networking.
+* `external` networking is done via a Daemonset
+
+TODO: Explain the difference between pod networking & inter-pod networking.

docs/security.md

@@ -0,0 +1,38 @@
+## Security Notes for Kubernetes
+
+## SSH Access
+
+SSH is allowed to the masters and the nodes, by default from anywhere.
+
+To change the CIDR allowed to access SSH (and HTTPS), set AdminAccess on the cluster spec.
+
+When using the default images, the SSH username will be `admin`, and the SSH private key is be
+the private key corresponding to the public key in `kops get secrets --type sshpublickey admin`.  When
+creating a new cluster, the SSH public key can be specified with the `--ssh-public-key` option, and it
+defaults to `~/.ssh/id_rsa.pub`.
+
+To change the SSH public key on an existing cluster:
+
+* `kops delete secret --name <clustername> sshpublickey admin`
+* `kops create secret --name <clustername> sshpublickey admin -i ~/.ssh/newkey.pub`
+* `kops update cluster --yes` to reconfigure the auto-scaling groups
+* `kops rolling-update --yes` to immediately roll all the machines so they have the new key (optional)
+
+
+## Kubernetes API
+
+(this section is a work in progress)
+
+Kubernetes has a number of authentication mechanisms:
+
+### API Bearer Token
+
+The API bearer token is a secret named 'admin'.
+
+`kops get secrets admin -oplaintext` will show it
+
+### Admin Access
+
+Access to the administrative API is stored in a secret named 'kube':
+
+`kops get secrets kube -oplaintext` or `kubectl config view --minify` to reveal

docs/terraform.md

@@ -0,0 +1,74 @@
+## Building Kubernetes clusters with terraform
+
+Kops can generate terraform configurations, and you can then apply them using the terraform plan/apply tools. 
+This is very handy if you are already using terraform, or if you want to check in the terraform output into
+version control.
+
+The terraform output should be reasonably stable (i.e. the text files should only change where something has actually
+changed - items should appear in the same order etc).
+
+
+### Using terraform
+
+To use terraform, you simple run update with `--target=terraform` (but see below for a workaround for a bug
+if you are using a terraform version before 0.7)
+
+For example, a complete setup might be:
+
+```
+export KOPS_STATE_STORE=s3://<somes3bucket>
+export CLUSTER_NAME=<kubernetes.mydomain.com>
+${GOPATH}/bin/kops create cluster ${NAME} --zones us-east-1c
+${GOPATH}/bin/kops update cluster ${NAME} --target=terraform
+
+cd out/terraform
+terraform plan
+terraform apply
+```
+
+When you eventually `terraform delete` the cluster, you should still run `kops delete cluster ${CLUSTER_NAME}`,
+to remove the kops cluster specification and any dynamically created Kubernetes resources (ELBs or volumes).
+
+### Workaround for Terraform versions before 0.7
+
+Before terraform version 0.7, there was a bug where it could not create AWS tags containing a dot.
+
+We recommend upgrading to version 0.7 or laster, which wil fix this bug.
+
+However, if you need to use an earlier version:
+
+This issue only affects the volumes.
+
+We divide the cloudup model into three parts:
+* models/config which contains all the options - this is run automatically by "create cluster"
+* models/proto which sets up the volumes and other data which would be hard to recover (e.g. likely keys & secrets in the near future)
+* models/cloudup which is the main cloud model for configuring everything else
+
+So the workaround is that you don't use terraform for the `proto` phase (you can't anyway, because of the bug!):
+
+```
+export KOPS_STATE_STORE=s3://<somes3bucket>
+export CLUSTER_NAME=<kubernetes.mydomain.com>
+${GOPATH}/bin/kops create cluster ${CLUSTER_NAME} --zones=us-east-1c
+${GOPATH}/bin/kops update cluster ${CLUSTER_NAME} --model=proto --yes
+```
+
+And then you can use terraform to do the remainder of the installation:
+
+```
+export CLUSTER_NAME=<kubernetes.mydomain.com>
+${GOPATH}/bin/kops update cluster ${CLUSTER_NAME} --model=cloudup --target=terraform
+```
+
+Then, to apply using terraform:
+
+```
+cd out/terraform
+
+terraform plan
+terraform apply
+```
+
+You should still run `kops delete cluster ${CLUSTER_NAME}`, to remove the kops cluster specification and any
+dynamically created Kubernetes resources (ELBs or volumes), but under this workaround also to remove the primary
+ELB volumes from the `proto` phase.

docs/testing.md

@@ -0,0 +1,48 @@
+## Testing tips
+
+If you are running kops as part of an e2e test, the following tips may be useful.
+
+### CI Kubernetes Build
+
+Set the KubernetesVersion to a `http://` or `https://` base url, such as `https://storage.googleapis.com/kubernetes-release-dev/ci/v1.4.0-alpha.2.677+ea69570f61af8e/`
+
+We expect the base url to have `bin/linux/amd64` directory containing:
+
+* kubelet
+* kubelet.sha1
+* kubectl
+* kubectl.sha1
+* kube-apiserver.docker_tag
+* kube-apiserver.tar
+* kube-apiserver.tar.sha1
+* kube-controller-manager.docker_tag
+* kube-controller-manager.tar
+* kube-controller-manager.tar.sha1
+* kube-proxy.docker_tag
+* kube-proxy.tar
+* kube-proxy.tar.sha1
+* kube-scheduler.docker_tag
+* kube-scheduler.tar
+* kube-scheduler.tar.sha1
+
+
+Do this with `kops edit cluster <clustername>`.  The spec should look like
+
+```
+...
+spec:
+  kubernetesVersion: "https://storage.googleapis.com/kubernetes-release-dev/ci/v1.4.0-alpha.2.677+ea69570f61af8e/"
+  cloudProvider: aws
+  etcdClusters:
+  - etcdMembers:
+    - name: us-east-1c
+      zone: us-east-1c
+    name: main
+...
+```
+
+
+### Running the kubernetes e2e test suite
+
+The [e2e](../e2e/README.md) directory has a docker image and some scripts which make it easy to run
+the kubernetes e2e tests, using kops.

docs/update_kops.md

@@ -0,0 +1,8 @@
+##How to update Kops - Kubernetes Ops
+
+Update the latest source code from kubernetes/kops
+
+```
+cd ${GOPATH}/src/k8s.io/kops/
+git pull && make
+```

docs/upgrade.md

@@ -0,0 +1,28 @@
+## Upgrading Kubernetes
+
+Upgrading Kubernetes is easy with kops.  The cluster spec contains a KubernetesVersion, so you
+can simply edit it with `kops edit`, and apply the updated configuration to your cluster.
+The `kops upgrade` command also automates checking for and applying updates.
+
+Note: if you want to upgrade from a `kube-up` installation, please see the instructions for
+[how to upgrade kubernetes installed with kube-up](upgrade_from_k8s_12.md).
+
+### Manual update
+
+* `kops edit cluster $NAME`
+* set the KubernetesVersion to the target version (e.g. `v1.3.5`)
+* `kops update cluster $NAME` to preview, then `kops update cluster $NAME --yes`
+* `kops rolling-update cluster $NAME` to preview, then `kops rolling-update cluster $NAME --yes`
+
+### Automated update
+
+* `kops upgrade cluster $NAME` to preview, then `kops upgrade cluster $NAME --yes`
+
+In future the upgrade step will likely perform the update immediately (and possibly even without a
+node restart), but currently you must:
+
+* `kops update cluster $NAME` to preview, then `kops update cluster $NAME --yes`
+* `kops rolling-update cluster $NAME` to preview, then `kops rolling-update cluster $NAME --yes`
+
+
+upgrade uses the latest Kuberentes stable release, published at `https://storage.googleapis.com/kubernetes-release/release/stable.txt`

docs/upgrade_from_k8s_12.md

@@ -44,7 +44,7 @@ Now have a look at the cluster configuration, to make sure it looks right.  If i
 open an issue.
 
 ```
-kops edit cluster ${OLD_NAME}
+kops get cluster ${OLD_NAME} -oyaml
 ````
 
 ## Move resources to a new cluster
@@ -62,7 +62,7 @@ The upgrade procedure forces you to choose a new cluster name (e.g. `k8s.mydomai
 
 ```
 export NEW_NAME=k8s.mydomain.com
-kops upgrade cluster --newname ${NEW_NAME} --name ${OLD_NAME}
+kops toolbox convert-imported --newname ${NEW_NAME} --name ${OLD_NAME}
 ```
 
 If you now list the clusters, you should see both the old cluster & the new cluster
@@ -73,6 +73,14 @@ kops get clusters
 
 You can also list the instance groups: `kops get ig --name ${NEW_NAME}`
 
+## Import the SSH public key
+
+The SSH public key is not easily retrieved from the old cluster, so you must add it:
+
+```
+kops create secret --name ${NEW_NAME} sshpublickey admin -i ~/.ssh/id_rsa.pub
+```
+
 ## Bring up the new cluster
 
 Use the update command to bring up the new cluster:
@@ -132,6 +140,10 @@ kubectl delete secret --namespace kube-system default-token-lhfkx
 Then restart the kube-dns pod so it picks up a valid secret:
 `kubectl delete pods --namespace kube-system --selector "k8s-app=kube-dns"`
 
+## Other fixes
+
+* If you're using a manually created ELB, the auto-scaling groups change, so you will need to reconfigure
+your ELBs to include the new auto-scaling group(s).
 
 ## Delete remaining resources of the old cluster
 

e2e/Dockerfile

@@ -0,0 +1,14 @@
+FROM debian:jessie
+
+RUN apt-get update && apt-get install --yes curl python-pip openssh-client
+RUN pip install awscli
+
+RUN curl https://sdk.cloud.google.com | bash
+
+RUN curl https://storage.googleapis.com/golang/go1.6.3.linux-amd64.tar.gz | tar -C /usr/local -xzf -
+
+ENV PATH /root/google-cloud-sdk/bin:/usr/local/go/bin:$PATH
+
+ADD runtests.sh /
+
+ADD conf /conf

e2e/Makefile

@@ -0,0 +1,22 @@
+#JOB_NAME=kubernetes-e2e-kops-aws
+#KUBERNETES_VERSION=v1.3.5
+#DNS_DOMAIN="mydomain.com"
+#JENKINS_GCS_LOGS_PATH=gs://kopeio-kubernetes-e2e/logs
+#KOPS_STATE_STORE=s3://mys3bucket
+
+test: image
+	docker run -v ${HOME}/.ssh/id_rsa.pub:/root/.ssh/id_rsa.pub \
+	           -v ${HOME}/.aws:/root/.aws \
+	           -v ${HOME}/.gsutil:/root/.gsutil \
+	           -v ${HOME}/.config/gcloud:/root/.config/gcloud \
+	           -e "DNS_DOMAIN=${DNS_DOMAIN}" \
+	           -e "JENKINS_GCS_LOGS_PATH=${JENKINS_GCS_LOGS_PATH}" \
+	           -e "KOPS_STATE_STORE=${KOPS_STATE_STORE}" \
+	           -e "JOB_NAME=${JOB_NAME}" \
+	           -e "KUBERNETES_VERSION=${KUBERNETES_VERSION}" \
+	           kubernetes-e2e-runner-kops \
+	           /runtests.sh
+
+image:
+	docker build -t kubernetes-e2e-runner-kops .
+

e2e/README.md

@@ -0,0 +1,18 @@
+## Run kubernetes e2e tests
+
+This docker image lets you run the kubernetes e2e tests very easily, using kops to create the cluster.
+
+You simply call make, specifying some variables that controls the build.
+
+An example:
+
+`make JOB_NAME=kubernetes-e2e-kops-aws KUBERNETES_VERSION=v1.3.5 DNS_DOMAIN=e2e.mydomain.com JENKINS_GCS_LOGS_PATH=gs://kopeio-kubernetes-e2e/logs KOPS_STATE_STORE=s3://clusters.mydomain.com`
+
+Variables:
+
+* `JOB_NAME` the e2e job to run.  Corresponds to a conf file in the conf directory.
+* `KUBERNETES_VERSION` the version of kubernetes to run.  Either a version like `v1.3.5`, or a URL prefix like `https://storage.googleapis.com/kubernetes-release-dev/ci/v1.4.0-alpha.2.677+ea69570f61af8e/`.  See [testing docs](../docs/testing.md)
+* `DNS_DOMAIN` the dns domain name to use for the cluster.  Must be a real domain name, with a zone registered in DNS (route53)
+* `JENKINS_GCS_LOGS_PATH` the gs bucket where we should upload the results of the build.  Note these will be publicly readable.
+* `KOPS_STATE_STORE` the url where the kops registry (store of cluster information) lives.
+

e2e/conf/cloud/aws

@@ -0,0 +1,12 @@
+##=============================================================
+# AWS Settings
+export AWS_DEFAULT_REGION=${AWS_REGION}
+#export KUBE_AWS_ZONE="us-west-2a"
+#export PROJECT="k8s-jkns-e2e-aws"
+export AWS_CONFIG_FILE=~/.aws/credentials
+export AWS_SSH_KEY=~/.ssh/kube_aws_rsa
+export KUBE_SSH_USER=admin
+# This is needed to be able to create PD from the e2e test
+export AWS_SHARED_CREDENTIALS_FILE=~/.aws/credentials
+
+export AWS_PROFILE=kubernetes-e2e

e2e/conf/cloud/gce

@@ -0,0 +1,5 @@
+# GCE variables
+#export INSTANCE_PREFIX="${E2E_NAME:-jenkins-e2e}"
+#export KUBE_GCE_NETWORK="${E2E_NAME:-jenkins-e2e}"
+#export KUBE_GCE_INSTANCE_PREFIX="${E2E_NAME:-jenkins-e2e}"
+#export GCE_SERVICE_ACCOUNT=$(gcloud auth list 2> /dev/null | grep active | cut -f3 -d' ')

e2e/conf/kubernetes-e2e-kops-aws

@@ -0,0 +1,12 @@
+export KUBERNETES_PROVIDER="aws"
+
+export E2E_MIN_STARTUP_PODS="1"
+export MASTER_SIZE="m3.large"
+export NODE_SIZE="m3.large"
+export NUM_NODES="3"
+export GINKGO_TEST_ARGS="--ginkgo.skip=\[Slow\]|\[Serial\]|\[Disruptive\]|\[Flaky\]|\[Feature:.+\]"
+export GINKGO_PARALLEL="y"
+
+# Central only has 2 AZs, so we use it for the non-HA test
+export NODE_ZONES="eu-central-1a"
+export AWS_REGION="eu-central-1"

e2e/conf/kubernetes-e2e-kops-aws-ha

@@ -0,0 +1,4 @@
+. conf/kubernetes-e2e-upup-aws
+
+export NODE_ZONES="eu-west-1a,eu-west-1b,eu-west-1c"
+export AWS_REGION="eu-west-1"

e2e/conf/site

@@ -0,0 +1,2 @@
+# Used to set site-specific settings
+# Most of these probably are better in the Makefile

e2e/runtests.sh

@@ -0,0 +1,215 @@
+#!/bin/bash -ex
+
+if [[ -z "${JOB_NAME}" ]]; then
+  echo "Must specify JOB_NAME env var"
+  exit 1
+fi
+if [[ -z "${KUBERNETES_VERSION}" ]]; then
+  echo "Must specify KUBERNETES_VERSION env var"
+  exit 1
+fi
+if [[ -z "${DNS_DOMAIN}" ]]; then
+  echo "Must specify DNS_DOMAIN env var"
+  exit 1
+fi
+if [[ -z "${KOPS_STATE_STORE}" ]]; then
+  echo "Must specify KOPS_STATE_STORE env var"
+  exit 1
+fi
+# TODO: Maybe skip if we don't want to upload logs?
+if [[ -z "${JENKINS_GCS_LOGS_PATH}" ]]; then
+  echo "Must specify JENKINS_GCS_LOGS_PATH env var"
+  exit 1
+fi
+
+echo "JOB_NAME=${JOB_NAME}"
+echo "Loading conf/${JOB_NAME}"
+
+. conf/${JOB_NAME}
+
+echo "Loading conf/cloud/${KUBERNETES_PROVIDER}"
+. conf/cloud/${KUBERNETES_PROVIDER}
+
+echo "Loading conf/site"
+. conf/site
+
+##=============================================================
+# Global settings
+export KUBE_GCS_RELEASE_BUCKET=kubernetes-release
+
+# We download the binaries ourselves
+# TODO: No way to tell e2e to use a particular release?
+# TODO: Maybe download and then bring up the cluster?
+export JENKINS_USE_EXISTING_BINARIES=y
+
+# This actually just skips kube-up master detection
+export KUBERNETES_CONFORMANCE_TEST=y
+
+##=============================================================
+# System settings (emulate jenkins)
+export USER=root
+export WORKSPACE=$HOME
+# Nothing should want Jenkins $HOME
+export HOME=${WORKSPACE}
+export BUILD_NUMBER=`date -u +%Y%m%d%H%M%S`
+export JENKINS_HOME=${HOME}
+
+# We'll directly up & down the cluster
+export E2E_UP="${E2E_UP:-false}"
+export E2E_TEST="${E2E_TEST:-true}"
+export E2E_DOWN="${E2E_DOWN:-false}"
+
+# Skip gcloud update checking
+export CLOUDSDK_COMPONENT_MANAGER_DISABLE_UPDATE_CHECK=true
+
+
+##=============================================================
+
+branch=master
+
+build_dir=${JENKINS_HOME}/jobs/${JOB_NAME}/builds/${BUILD_NUMBER}/
+rm -rf ${build_dir}
+mkdir -p ${build_dir}/workspace
+
+cd ${build_dir}/workspace
+
+# Sanity check
+#gsutil ls ${JENKINS_GCS_LOGS_PATH}
+
+exit_code=0
+SECONDS=0 # magic bash timer variable
+curl -fsS --retry 3  "https://raw.githubusercontent.com/kubernetes/kubernetes/master/hack/jenkins/e2e-runner.sh" > /tmp/e2e.sh
+chmod +x /tmp/e2e.sh
+
+# We need kubectl to write kubecfg from kops
+curl -fsS --retry 3  "https://storage.googleapis.com/kubernetes-release/release/v1.3.5/bin/linux/amd64/kubectl" > /usr/local/bin/kubectl
+chmod +x /usr/local/bin/kubectl
+
+curl -fsS --retry 3  "https://kubeupv2.s3.amazonaws.com/kops/kops-1.3.tar.gz" > /tmp/kops.tar.gz
+tar zxf /tmp/kops.tar.gz -C /opt
+
+if [[ ! -e ${AWS_SSH_KEY} ]]; then
+  echo "Creating ssh key ${AWS_SSH_KEY}"
+  ssh-keygen -N "" -t rsa -f ${AWS_SSH_KEY}
+fi
+
+function fetch_tars_from_gcs() {
+    local -r bucket="${1}"
+    local -r build_version="${2}"
+    echo "Pulling binaries from GCS; using server version ${bucket}/${build_version}."
+    gsutil -mq cp \
+        "gs://${KUBE_GCS_RELEASE_BUCKET}/${bucket}/${build_version}/kubernetes.tar.gz" \
+        "gs://${KUBE_GCS_RELEASE_BUCKET}/${bucket}/${build_version}/kubernetes-test.tar.gz" \
+        .
+}
+
+function unpack_binaries() {
+    md5sum kubernetes*.tar.gz
+    tar -xzf kubernetes.tar.gz
+    tar -xzf kubernetes-test.tar.gz
+}
+
+
+fetch_tars_from_gcs release ${KUBERNETES_VERSION}
+unpack_binaries
+
+# Clean up everything when we're done
+function finish {
+  /opt/kops/kops delete cluster \
+                 --name ${JOB_NAME}.${DNS_DOMAIN} \
+                 --yes  2>&1 | tee -a ${build_dir}/build-log.txt
+}
+trap finish EXIT
+
+set -e
+
+# Create the cluster spec
+pushd /opt/kops
+/opt/kops/kops create cluster \
+                  --name ${JOB_NAME}.${DNS_DOMAIN} \
+                  --cloud ${KUBERNETES_PROVIDER} \
+                  --zones ${NODE_ZONES} \
+                  --node-size ${NODE_SIZE} \
+                  --master-size ${MASTER_SIZE} \
+                  --ssh-public-key ${AWS_SSH_KEY}.pub \
+                  --kubernetes-version ${KUBERNETES_VERSION} \
+                  --v=4 2>&1 | tee -a ${build_dir}/build-log.txt
+exit_code=${PIPESTATUS[0]}
+popd
+
+# Apply the cluster spec
+if [[ ${exit_code} == 0 ]]; then
+  pushd /opt/kops
+  /opt/kops/kops update cluster \
+                    --name ${JOB_NAME}.${DNS_DOMAIN} \
+                    --yes --v=4 2>&1 | tee -a ${build_dir}/build-log.txt
+  exit_code=${PIPESTATUS[0]}
+  popd
+fi
+
+# Wait for kubectl to begin responding (at least master up)
+if [[ ${exit_code} == 0 ]]; then
+  attempt=0
+  while true; do
+    kubectl get nodes --show-labels  2>&1 | tee -a ${build_dir}/build-log.txt
+    exit_code=${PIPESTATUS[0]}
+
+    if [[ ${exit_code} == 0 ]]; then
+      break
+    fi
+    if (( attempt > 60 )); then
+      echo "Unable to connect to API in 15 minutes (master did not launch?)"
+      break
+    fi
+    attempt=$(($attempt+1))
+    sleep 15
+  done
+fi
+
+# TODO: can we get rid of this?
+echo "API responded; waiting 450 seconds for DNS to settle"
+for ((i=1;i<=15;i++)); do
+    kubectl get nodes --show-labels  2>&1 | tee -a ${build_dir}/build-log.txt
+    sleep 30
+done
+
+
+# Run e2e tests
+if [[ ${exit_code} == 0 ]]; then
+  /tmp/e2e.sh 2>&1 | tee -a ${build_dir}/build-log.txt
+  exit_code=${PIPESTATUS[0]}
+fi
+
+# Try to clean up normally so it goes into the logs
+# (we have an exit hook for abnormal termination, but that does not get logged)
+finish
+
+duration=$SECONDS
+set +e
+
+if [[ ${exit_code} == 0 ]]; then
+  success="true"
+else
+  success="false"
+fi
+
+version=`cat kubernetes/version`
+
+gcs_acl="public-read"
+gcs_job_path="${JENKINS_GCS_LOGS_PATH}/${JOB_NAME}"
+gcs_build_path="${gcs_job_path}/${BUILD_NUMBER}"
+
+gsutil -q cp -a "${gcs_acl}" -z txt "${build_dir}/build-log.txt" "${gcs_build_path}/"
+
+curl -fsS --retry 3 "https://raw.githubusercontent.com/kubernetes/kubernetes/master/hack/jenkins/upload-to-gcs.sh" | bash -
+
+
+curl -fsS --retry 3 "https://raw.githubusercontent.com/kubernetes/kubernetes/master/hack/jenkins/upload-finished.sh" > upload-finished.sh
+chmod +x upload-finished.sh
+
+if [[ ${exit_code} == 0 ]]; then
+  ./upload-finished.sh SUCCESS
+else
+  ./upload-finished.sh UNSTABLE
+fi
+

images/nodeup-builder/Dockerfile

@@ -0,0 +1,13 @@
+FROM debian:jessie
+
+# Install packages:
+#  curl (to download golang)
+#  git (for getting the current head)
+#  gcc make (for compilation)
+RUN apt-get update && apt-get install --yes curl git gcc make
+
+# Install golang
+RUN curl -L https://storage.googleapis.com/golang/go1.6.3.linux-amd64.tar.gz | tar zx -C /usr/local
+ENV PATH $PATH:/usr/local/go/bin
+
+COPY images/nodeup-builder/onbuild.sh /onbuild.sh

images/nodeup-builder/onbuild.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+mkdir -p /go
+export GOPATH=/go
+
+mkdir -p /go/src/k8s.io
+ln -s /src/ /go/src/k8s.io/kops
+
+cd /go/src/k8s.io/kops/
+make nodeup-gocode
+
+mkdir -p /src/.build/artifacts/
+cp /go/bin/nodeup /src/.build/artifacts/

images/protokube-builder/Dockerfile

@@ -7,7 +7,7 @@ FROM debian:jessie
 RUN apt-get update && apt-get install --yes curl git gcc make
 
 # Install golang
-RUN curl -L https://storage.googleapis.com/golang/go1.6.2.linux-amd64.tar.gz | tar zx -C /usr/local
+RUN curl -L https://storage.googleapis.com/golang/go1.6.3.linux-amd64.tar.gz | tar zx -C /usr/local
 ENV PATH $PATH:/usr/local/go/bin
 
 COPY images/protokube-builder/onbuild.sh /onbuild.sh

upup/models/cloudup/_aws/master/_master_asg/master_asg.yaml

@@ -2,7 +2,7 @@
 
 # ASG for master
 launchConfiguration/{{ $m.Name }}.masters.{{ ClusterName }}:
-  sshKey: sshKey/{{ ClusterName }}
+  sshKey: sshKey/{{ SSHKeyName }}
   securityGroups:
     - securityGroup/masters.{{ ClusterName }}
   iamInstanceProfile: iamInstanceProfile/masters.{{ ClusterName }}

upup/models/cloudup/_aws/master/_master_single/master_single.yaml

@@ -28,7 +28,7 @@ instance/master.{{ ClusterName }}:
 {{ if not (HasTag "_master_lb") }}
     k8s.io/dns/public: "api.{{ ClusterName }}"
 {{ end }}
-  sshKey: sshKey/{{ ClusterName }}
+  sshKey: sshKey/{{ SSHKeyName }}
   securityGroups:
     - securityGroup/master.{{ ClusterName }}
   iamInstanceProfile: iamInstanceProfile/master.{{ ClusterName }}

upup/models/cloudup/_aws/nodes.yaml

@@ -46,7 +46,7 @@ securityGroupRule/all-node-to-master:
 
 # LaunchConfiguration & ASG for nodes
 launchConfiguration/{{ $nodeset.Name }}.{{ ClusterName }}:
-  sshKey: sshKey/{{ ClusterName }}
+  sshKey: sshKey/{{ SSHKeyName }}
   securityGroups:
     - securityGroup/nodes.{{ ClusterName }}
   iamInstanceProfile: iamInstanceProfile/nodes.{{ ClusterName }}

upup/models/cloudup/_aws/resources/nodeup.sh.template

@@ -113,7 +113,7 @@ echo "== nodeup node config starting =="
 ensure-install-dir
 
 cat > kube_env.yaml << __EOF_KUBE_ENV
-{{ RenderResource "resources/config.yaml" Args }}
+{{ RenderNodeUpConfig Args }}
 __EOF_KUBE_ENV
 
 download-release

upup/models/cloudup/_aws/sshkey.yaml

@@ -1,3 +1,2 @@
-sshKey/{{ ClusterName }}:
+sshKey/{{ SSHKeyName}}:
-  name: kubernetes.{{ClusterName}}
   publicKey: resources/ssh-public-key

upup/models/cloudup/resources/config.yaml.template

@@ -1,13 +1 @@
-Tags:
+{{ RenderNodeUpConfig Args }}
-{{ range $tag := ComputeNodeTags Args }}
-  - {{ $tag }}
-{{ end }}
-
-
-Assets:
-{{ range $asset := Assets }}
-  - {{ $asset }}
-{{ end }}
-
-ClusterName: {{ ClusterName }}
-ClusterLocation: {{ ClusterLocation }}

upup/models/config/components/docker/_kubenet/kubenet.options

@@ -1,5 +0,0 @@
-{% set log_level = "--log-level=warn" -%}
-{% if pillar['docker_test_log_level'] is defined -%}
-  {% set log_level = pillar['docker_test_log_level'] -%}
-{% endif -%}
-docker.bridge=

upup/models/config/components/docker/_networking_kubenet/kubenet.options

@@ -0,0 +1,2 @@
+Docker:
+  Bridge: ""

upup/models/config/components/kube-apiserver/kube-apiserver.options

@@ -14,4 +14,4 @@ KubeAPIServer:
   TokenAuthFile: /srv/kubernetes/known_tokens.csv
   LogLevel: 2
   AllowPrivileged: true
-  Image: gcr.io/google_containers/kube-apiserver:v{{ .KubernetesVersion }}
+  Image: {{ Image "kube-apiserver" }}

upup/models/config/components/kube-controller-manager/_networking_classic/kube-controller-manager.networking-classic.options

@@ -0,0 +1,2 @@
+KubeControllerManager:
+  ConfigureCloudRoutes: true

upup/models/config/components/kube-controller-manager/_networking_kubenet/_networking_external/kube-controller-manager.networking-external.options

@@ -0,0 +1,2 @@
+KubeControllerManager:
+  ConfigureCloudRoutes: false

upup/models/config/components/kube-controller-manager/_networking_kubenet/kube-controller-manager.networking-kubenet.options

@@ -0,0 +1,2 @@
+KubeControllerManager:
+  ConfigureCloudRoutes: true

upup/models/config/components/kube-controller-manager/kube-controller-manager.options

@@ -2,11 +2,12 @@ KubeControllerManager:
   PathSrvKubernetes: /srv/kubernetes
   Master: 127.0.0.1:8080
   AllocateNodeCIDRs: true
+  ConfigureCloudRoutes: false
   ServiceAccountPrivateKeyFile: /srv/kubernetes/server.key
   LogLevel: 2
   RootCAFile: /srv/kubernetes/ca.crt
   ClusterName: {{ ClusterName }}
-  Image: gcr.io/google_containers/kube-controller-manager:v{{ .KubernetesVersion }}
+  Image: {{ Image "kube-controller-manager" }}
   # Doesn't seem to be any real downside to always doing a leader election
   LeaderElection:
     LeaderElect: true

upup/models/config/components/kube-dns/kube-dns.options

@@ -1,4 +1,6 @@
 KubeDNS:
-  Replicas: 1
+  Replicas: 2
   ServerIP: {{ WellKnownServiceIP 10 }}
   Domain: {{ .ClusterDNSDomain }}
+  # TODO: Once we start shipping more images, start using them
+  Image: gcr.io/google_containers/kubedns-amd64:1.3

upup/models/config/components/kube-proxy/kube-proxy.options

@@ -7,6 +7,6 @@ KubeProxy:
   # requests of other per-node add-ons (e.g. fluentd).
   CPURequest: 20m
 
-  Image: gcr.io/google_containers/kube-proxy:v{{ .KubernetesVersion }}
+  Image: {{ Image "kube-proxy" }}
 
   Master: https://{{ .MasterInternalName }}

upup/models/config/components/kube-scheduler/kube-scheduler.options

@@ -1,7 +1,7 @@
 KubeScheduler:
   Master: 127.0.0.1:8080
   LogLevel: 2
-  Image: gcr.io/google_containers/kube-scheduler:v{{ .KubernetesVersion }}
+  Image: {{ Image "kube-scheduler" }}
   # Doesn't seem to be any real downside to always doing a leader election
   LeaderElection:
     LeaderElect: true

upup/models/config/components/kubelet/_networking_kubenet/kubelet.kubenet.options

@@ -0,0 +1,3 @@
+Kubelet:
+  NetworkPluginName: kubenet
+  ReconcileCIDR: true

upup/models/nodeup/_protokube/services/protokube.service.template

@@ -5,8 +5,8 @@ After=docker.service
 
 [Service]
 EnvironmentFile=/etc/sysconfig/protokube
-ExecStartPre=/usr/bin/docker pull kope/protokube:1.3
+ExecStartPre=/usr/bin/docker pull {{ ProtokubeImage }}
-ExecStart=/usr/bin/docker run -v /:/rootfs/ --privileged kope/protokube:1.3 /usr/bin/protokube "$DAEMON_ARGS"
+ExecStart=/usr/bin/docker run -v /:/rootfs/ --privileged {{ ProtokubeImage }} /usr/bin/protokube "$DAEMON_ARGS"
 Restart=always
 RestartSec=2s
 StartLimitInterval=0

upup/models/nodeup/docker/_systemd/services/docker.service.meta

@@ -1,3 +0,0 @@
-{
-  "manageState": false
-}

upup/models/nodeup/networking/_cni_bridge/files/opt/cni/bin/bridge.asset

@@ -0,0 +1,2 @@
+{
+}

upup/models/nodeup/networking/_cni_bridge/files/opt/cni/bin/bridge.asset.meta

@@ -0,0 +1,3 @@
+{
+  "mode": "0755"
+}

upup/models/nodeup/networking/_cni_host_local/files/opt/cni/bin/host-local.asset

@@ -0,0 +1,2 @@
+{
+}

upup/models/nodeup/networking/_cni_host_local/files/opt/cni/bin/host-local.asset.meta

@@ -0,0 +1,3 @@
+{
+  "mode": "0755"
+}

upup/models/nodeup/networking/_cni_loopback/files/opt/cni/bin/loopback.asset

@@ -0,0 +1,2 @@
+{
+}

upup/models/nodeup/networking/_cni_loopback/files/opt/cni/bin/loopback.asset.meta

@@ -0,0 +1,3 @@
+{
+  "mode": "0755"
+}

upup/models/nodeup/networking/_cni_ptp/files/opt/cni/bin/ptp.asset

@@ -0,0 +1,2 @@
+{
+}

upup/models/nodeup/networking/_cni_ptp/files/opt/cni/bin/ptp.asset.meta

@@ -0,0 +1,3 @@
+{
+  "mode": "0755"
+}

upup/pkg/api/cluster.go

@@ -4,6 +4,7 @@ import (
 	"encoding/binary"
 	"fmt"
 	"github.com/golang/glog"
+	"k8s.io/kops/upup/pkg/fi/vfs"
 	k8sapi "k8s.io/kubernetes/pkg/api"
 	"k8s.io/kubernetes/pkg/api/unversioned"
 	"net"
@@ -76,8 +77,6 @@ type ClusterSpec struct {
 	// ClusterName is a unique identifier for the cluster, and currently must be a DNS name
 	//ClusterName       string `json:",omitempty"`
 
-	//AllocateNodeCIDRs *bool `json:"allocateNodeCIDRs,omitempty"`
-
 	Multizone *bool `json:"multizone,omitempty"`
 
 	//ClusterIPRange                string `json:",omitempty"`
@@ -103,8 +102,6 @@ type ClusterSpec struct {
 	//  * enable debugging handlers on the master, so kubectl logs works
 	IsolateMasters *bool `json:"isolateMasters,omitempty"`
 
-	//NetworkProvider               string `json:",omitempty"`
-	//
 	//HairpinMode                   string `json:",omitempty"`
 	//
 	//OpencontrailTag               string `json:",omitempty"`
@@ -209,9 +206,15 @@ type ClusterSpec struct {
 	KubeProxy             *KubeProxyConfig             `json:"kubeProxy,omitempty"`
 	Kubelet               *KubeletConfig               `json:"kubelet,omitempty"`
 	MasterKubelet         *KubeletConfig               `json:"masterKubelet,omitempty"`
+
+	// Networking configuration
+	Networking *NetworkingSpec `json:"networking,omitempty"`
 }
 
 type KubeDNSConfig struct {
+	// Image is the name of the docker image to run
+	Image string `json:"image,omitempty"`
+
 	Replicas int    `json:"replicas,omitempty"`
 	Domain   string `json:"domain,omitempty"`
 	ServerIP string `json:"serverIP,omitempty"`
@@ -265,6 +268,7 @@ type ClusterZoneSpec struct {
 // PerformAssignments populates values that are required and immutable
 // For example, it assigns stable Keys to NodeSets & Masters, and
 // it assigns CIDRs to subnets
+// We also assign KubernetesVersion, because we want it to be explicit
 func (c *Cluster) PerformAssignments() error {
 	if c.Spec.NetworkCIDR == "" && !c.SharedVPC() {
 		// TODO: Choose non-overlapping networking CIDRs for VPCs?
@@ -286,6 +290,11 @@ func (c *Cluster) PerformAssignments() error {
 		}
 	}
 
+	err := c.ensureKubernetesVersion()
+	if err != nil {
+		return err
+	}
+
 	return nil
 }
 
@@ -297,9 +306,55 @@ func (c *Cluster) FillDefaults() error {
 		c.Spec.AdminAccess = append(c.Spec.AdminAccess, "0.0.0.0/0")
 	}
 
+	if c.Spec.Networking == nil {
+		c.Spec.Networking = &NetworkingSpec{}
+	}
+
+	if c.Spec.Networking.Classic != nil {
+		// OK
+	} else if c.Spec.Networking.Kubenet != nil {
+		// OK
+	} else if c.Spec.Networking.External != nil {
+		// OK
+	} else {
+		// No networking model selected; choose Classic
+		c.Spec.Networking.Classic = &ClassicNetworkingSpec{}
+	}
+
+	err := c.ensureKubernetesVersion()
+	if err != nil {
+		return err
+	}
+
 	return nil
 }
 
+// ensureKubernetesVersion populates KubernetesVersion, if it is not already set
+// It will be populated with the latest stable kubernetes version
+func (c *Cluster) ensureKubernetesVersion() error {
+	if c.Spec.KubernetesVersion == "" {
+		latestVersion, err := FindLatestKubernetesVersion()
+		if err != nil {
+			return err
+		}
+		glog.Infof("Using kubernetes latest stable version: %s", latestVersion)
+		c.Spec.KubernetesVersion = latestVersion
+	}
+	return nil
+}
+
+// FindLatestKubernetesVersion returns the latest kubernetes version,
+// as stored at https://storage.googleapis.com/kubernetes-release/release/stable.txt
+func FindLatestKubernetesVersion() (string, error) {
+	stableURL := "https://storage.googleapis.com/kubernetes-release/release/stable.txt"
+	b, err := vfs.Context.ReadFile(stableURL)
+	if err != nil {
+		return "", fmt.Errorf("KubernetesVersion not specified, and unable to download latest version from %q: %v", stableURL, err)
+	}
+	latestVersion := strings.TrimSpace(string(b))
+	return latestVersion, nil
+}
+
 func (z *ClusterZoneSpec) performAssignments(c *Cluster) error {
 	if z.CIDR == "" {
 		cidr, err := z.assignCIDR(c)

upup/pkg/api/componentconfig.go

@@ -147,9 +147,9 @@ type KubeletConfig struct {
 	//LowDiskSpaceThresholdMB int32 `json:"lowDiskSpaceThresholdMB"`
 	//// How frequently to calculate and cache volume disk usage for all pods
 	//VolumeStatsAggPeriod unversioned.Duration `json:"volumeStatsAggPeriod"`
-	//// networkPluginName is the name of the network plugin to be invoked for
+	// networkPluginName is the name of the network plugin to be invoked for
-	//// various events in kubelet/pod lifecycle
+	// various events in kubelet/pod lifecycle
-	//NetworkPluginName string `json:"networkPluginName"`
+	NetworkPluginName string `json:"networkPluginName,omitempty" flag:"network-plugin"`
 	//// networkPluginDir is the full path of the directory in which to search
 	//// for network plugins
 	//NetworkPluginDir string `json:"networkPluginDir"`
@@ -337,11 +337,13 @@ type KubeProxyConfig struct {
 }
 
 type DockerConfig struct {
-	Bridge   string `json:"bridge,omitempty" flag:"bridge"`
+	Bridge           string `json:"bridge,omitempty" flag:"bridge"`
-	LogLevel string `json:"logLevel,omitempty" flag:"log-level"`
+	LogLevel         string `json:"logLevel,omitempty" flag:"log-level"`
-	IPTables bool   `json:"ipTables,omitempty" flag:"iptables"`
+	IPTables         *bool  `json:"ipTables,omitempty" flag:"iptables"`
-	IPMasq   bool   `json:"ipMasq,omitempty" flag:"ip-masq"`
+	IPMasq           *bool  `json:"ipMasq,omitempty" flag:"ip-masq"`
-	Storage  string `json:"storage,omitempty" flag:"s"`
+	Storage          string `json:"storage,omitempty" flag:"storage-driver"`
+	InsecureRegistry string `json:"insecureRegistry,omitempty" flag:"insecure-registry"`
+	MTU              *int   `json:"mtu,omitempty" flag:"mtu"`
 }
 
 type KubeAPIServerConfig struct {

upup/pkg/api/labels.go

@@ -0,0 +1,7 @@
+package api
+
+// AnnotationNameManagement is the annotation that indicates that a cluster is under external or non-standard management
+const AnnotationNameManagement = "kops.kubernetes.io/management"
+
+// AnnotationValueManagementImported is the annotation value that indicates a cluster was imported, typically as part of an upgrade
+const AnnotationValueManagementImported = "imported"

upup/pkg/api/networking.go

@@ -0,0 +1,22 @@
+package api
+
+// NetworkingSpec allows selection and configuration of a networking plugin
+type NetworkingSpec struct {
+	Classic  *ClassicNetworkingSpec  `json:"classic,omitempty"`
+	Kubenet  *KubenetNetworkingSpec  `json:"kubenet,omitempty"`
+	External *ExternalNetworkingSpec `json:"external,omitempty"`
+}
+
+// ClassicNetworkingSpec is the specification of classic networking mode, integrated into kubernetes
+type ClassicNetworkingSpec struct {
+}
+
+// KubenetNetworkingSpec is the specification for kubenet networking, largely integrated but intended to replace classic
+type KubenetNetworkingSpec struct {
+}
+
+// ExternalNetworkingSpec is the specification for networking that is implemented by a Daemonset
+// Networking is not managed by kops - we can create options here that directly configure e.g. weave
+// but this is useful for arbitrary network modes or for modes that don't need additional configuration.
+type ExternalNetworkingSpec struct {
+}

upup/pkg/api/validation.go

@@ -2,8 +2,10 @@ package api
 
 import (
 	"fmt"
+	"k8s.io/kubernetes/pkg/util/validation"
 	"net"
 	"net/url"
+	"strings"
 )
 
 func (c *Cluster) Validate(strict bool) error {
@@ -13,6 +15,18 @@ func (c *Cluster) Validate(strict bool) error {
 		return fmt.Errorf("Cluster Name is required (e.g. --name=mycluster.myzone.com)")
 	}
 
+	{
+		// Must be a dns name
+		errs := validation.IsDNS1123Subdomain(c.Name)
+		if len(errs) != 0 {
+			return fmt.Errorf("Cluster Name must be a valid DNS name (e.g. --name=mycluster.myzone.com) errors: %s", strings.Join(errs, ", "))
+		}
+
+		if !strings.Contains(c.Name, ".") {
+			return fmt.Errorf("Cluster Name must be a fully-qualified DNS name (e.g. --name=mycluster.myzone.com)")
+		}
+	}
+
 	if len(c.Spec.Zones) == 0 {
 		// TODO: Auto choose zones from region?
 		return fmt.Errorf("must configure at least one Zone (use --zones)")

upup/pkg/api/validation_test.go

@@ -0,0 +1,15 @@
+package api
+
+import (
+	"k8s.io/kubernetes/pkg/util/validation"
+	"testing"
+)
+
+func Test_Validate_DNS(t *testing.T) {
+	for _, name := range []string{"test.-", "!", "-"} {
+		errs := validation.IsDNS1123Subdomain(name)
+		if len(errs) == 0 {
+			t.Fatalf("Expected errors validating name %q", name)
+		}
+	}
+}

upup/pkg/fi/assetstore.go

@@ -69,13 +69,13 @@ func (r *assetResource) GetSource() *Source {
 }
 
 type AssetStore struct {
-	assetDir string
+	cacheDir string
 	assets   []*asset
 }
 
-func NewAssetStore(assetDir string) *AssetStore {
+func NewAssetStore(cacheDir string) *AssetStore {
 	a := &AssetStore{
-		assetDir: assetDir,
+		cacheDir: cacheDir,
 	}
 	return a
 }
@@ -163,7 +163,7 @@ func (a *AssetStore) addURL(url string, hash *hashing.Hash) error {
 		}
 	}
 
-	localFile := path.Join(a.assetDir, hash.String()+"_"+utils.SanitizeString(url))
+	localFile := path.Join(a.cacheDir, hash.String()+"_"+utils.SanitizeString(url))
 	_, err = DownloadURL(url, localFile, hash)
 	if err != nil {
 		return err
@@ -237,7 +237,7 @@ func (a *AssetStore) addURL(url string, hash *hashing.Hash) error {
 //}
 
 func (a *AssetStore) addArchive(archiveSource *Source, archiveFile string) error {
-	extracted := path.Join(a.assetDir, "extracted/"+path.Base(archiveFile))
+	extracted := path.Join(a.cacheDir, "extracted/"+path.Base(archiveFile))
 
 	// TODO: Use a temp file so this is atomic
 	if _, err := os.Stat(extracted); os.IsNotExist(err) {

upup/pkg/fi/ca.go

@@ -32,6 +32,9 @@ const (
 	SecretTypeSSHPublicKey = "SSHPublicKey"
 	SecretTypeKeypair      = "Keypair"
 	SecretTypeSecret       = "Secret"
+
+	// Name for the primary SSH key
+	SecretNameSSHPrimary = "admin"
 )
 
 type KeystoreItem struct {

upup/pkg/fi/cloudup/apply_cluster.go

@@ -3,7 +3,6 @@ package cloudup
 import (
 	"fmt"
 	"github.com/golang/glog"
-	"io/ioutil"
 	"k8s.io/kops/upup/pkg/api"
 	"k8s.io/kops/upup/pkg/fi"
 	"k8s.io/kops/upup/pkg/fi/cloudup/awstasks"
@@ -12,11 +11,16 @@ import (
 	"k8s.io/kops/upup/pkg/fi/cloudup/gcetasks"
 	"k8s.io/kops/upup/pkg/fi/cloudup/terraform"
 	"k8s.io/kops/upup/pkg/fi/fitasks"
+	"k8s.io/kops/upup/pkg/fi/hashing"
+	"k8s.io/kops/upup/pkg/fi/nodeup"
+	"k8s.io/kops/upup/pkg/fi/vfs"
 	"os"
 	"path"
 	"strings"
 )
 
+const MaxAttemptsWithNoProgress = 3
+
 type ApplyClusterCmd struct {
 	Cluster *api.Cluster
 
@@ -28,12 +32,12 @@ type ApplyClusterCmd struct {
 	// Models is a list of cloudup models to apply
 	Models []string
 
-	// Target specifies how we are operating e.g. direct to GCE, or AWS, or dry-run, or terraform
+	// TargetName specifies how we are operating e.g. direct to GCE, or AWS, or dry-run, or terraform
-	Target string
+	TargetName string
-	//// The node model to use
+
-	//NodeModel string
+	// Target is the fi.Target we will operate against
-	// The SSH public key (file) to use
+	Target fi.Target
-	SSHPublicKey string
+
 	// OutDir is a local directory in which we place output, can cache files etc
 	OutDir string
 
@@ -56,10 +60,6 @@ func (c *ApplyClusterCmd) Run() error {
 		return err
 	}
 
-	// TODO: Make these configurable?
-	useMasterASG := true
-	useMasterLB := false
-
 	err = api.DeepValidate(c.Cluster, c.InstanceGroups, true)
 	if err != nil {
 		return err
@@ -78,8 +78,6 @@ func (c *ApplyClusterCmd) Run() error {
 		return fmt.Errorf("ClusterRegistry is required")
 	}
 
-	tags := make(map[string]struct{})
-
 	l := &Loader{}
 	l.Init()
 	l.Cluster = c.Cluster
@@ -99,8 +97,16 @@ func (c *ApplyClusterCmd) Run() error {
 	}
 
 	if len(c.Assets) == 0 {
+		var baseURL string
+		if isBaseURL(cluster.Spec.KubernetesVersion) {
+			baseURL = cluster.Spec.KubernetesVersion
+		} else {
+			baseURL = "https://storage.googleapis.com/kubernetes-release/release/v" + cluster.Spec.KubernetesVersion
+		}
+		baseURL = strings.TrimSuffix(baseURL, "/")
+
 		{
-			defaultKubeletAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/bin/linux/amd64/kubelet", cluster.Spec.KubernetesVersion)
+			defaultKubeletAsset := baseURL + "/bin/linux/amd64/kubelet"
 			glog.Infof("Adding default kubelet release asset: %s", defaultKubeletAsset)
 
 			hash, err := findHash(defaultKubeletAsset)
@@ -111,7 +117,7 @@ func (c *ApplyClusterCmd) Run() error {
 		}
 
 		{
-			defaultKubectlAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/release/v%s/bin/linux/amd64/kubectl", cluster.Spec.KubernetesVersion)
+			defaultKubectlAsset := baseURL + "/bin/linux/amd64/kubectl"
 			glog.Infof("Adding default kubectl release asset: %s", defaultKubectlAsset)
 
 			hash, err := findHash(defaultKubectlAsset)
@@ -120,39 +126,34 @@ func (c *ApplyClusterCmd) Run() error {
 			}
 			c.Assets = append(c.Assets, hash.Hex()+"@"+defaultKubectlAsset)
 		}
+
+		if usesCNI(cluster) {
+			defaultCNIAsset := fmt.Sprintf("https://storage.googleapis.com/kubernetes-release/network-plugins/cni-8a936732094c0941e1543ef5d292a1f4fffa1ac5.tar.gz")
+			glog.Infof("Adding default CNI asset: %s", defaultCNIAsset)
+
+			hashString := "86966c78cc9265ee23f7892c5cad0ec7590cec93"
+			//hash, err := findHash(defaultCNIAsset)
+			//if err != nil {
+			//	return err
+			//}
+			//hashString := hash.Hex()
+			c.Assets = append(c.Assets, hashString+"@"+defaultCNIAsset)
+		}
 	}
 
 	if c.NodeUpSource == "" {
-		location := "https://kubeupv2.s3.amazonaws.com/nodeup/nodeup-1.3.tar.gz"
+		location := os.Getenv("NODEUP_URL")
-		glog.Infof("Using default nodeup location: %q", location)
+		if location == "" {
+			location = "https://kubeupv2.s3.amazonaws.com/nodeup/nodeup-1.3.tar.gz"
+			glog.Infof("Using default nodeup location: %q", location)
+		} else {
+			glog.Warningf("Using nodeup location from NODEUP_URL env var: %q", location)
+		}
 		c.NodeUpSource = location
 	}
 
 	checkExisting := true
 
-	var nodeUpTags []string
-	nodeUpTags = append(nodeUpTags, "_protokube")
-
-	if useMasterASG {
-		tags["_master_asg"] = struct{}{}
-	} else {
-		tags["_master_single"] = struct{}{}
-	}
-
-	if useMasterLB {
-		tags["_master_lb"] = struct{}{}
-	} else {
-		tags["_not_master_lb"] = struct{}{}
-	}
-
-	if cluster.Spec.MasterPublicName != "" {
-		tags["_master_dns"] = struct{}{}
-	}
-
-	if fi.BoolValue(cluster.Spec.IsolateMasters) {
-		tags["_isolate_masters"] = struct{}{}
-	}
-
 	l.AddTypes(map[string]interface{}{
 		"keypair": &fitasks.Keypair{},
 		"secret":  &fitasks.Secret{},
@@ -166,6 +167,18 @@ func (c *ApplyClusterCmd) Run() error {
 	region := ""
 	project := ""
 
+	var sshPublicKeys [][]byte
+	{
+		keys, err := keyStore.FindSSHPublicKeys(fi.SecretNameSSHPrimary)
+		if err != nil {
+			return fmt.Errorf("error retrieving SSH public key %q: %v", fi.SecretNameSSHPrimary, err)
+		}
+
+		for _, k := range keys {
+			sshPublicKeys = append(sshPublicKeys, k.Data)
+		}
+	}
+
 	switch cluster.Spec.CloudProvider {
 	case "gce":
 		{
@@ -174,8 +187,6 @@ func (c *ApplyClusterCmd) Run() error {
 			project = gceCloud.Project
 
 			glog.Fatalf("GCE is (probably) not working currently - please ping @justinsb for cleanup")
-			tags["_gce"] = struct{}{}
-			nodeUpTags = append(nodeUpTags, "_gce")
 
 			l.AddTypes(map[string]interface{}{
 				"persistentDisk":       &gcetasks.PersistentDisk{},
@@ -193,9 +204,6 @@ func (c *ApplyClusterCmd) Run() error {
 			awsCloud := cloud.(*awsup.AWSCloud)
 			region = awsCloud.Region
 
-			tags["_aws"] = struct{}{}
-			nodeUpTags = append(nodeUpTags, "_aws")
-
 			l.AddTypes(map[string]interface{}{
 				// EC2
 				"elasticIP":                   &awstasks.ElasticIP{},
@@ -237,8 +245,25 @@ func (c *ApplyClusterCmd) Run() error {
 				"dnsZone": &awstasks.DNSZone{},
 			})
 
-			if c.SSHPublicKey == "" {
+			if len(sshPublicKeys) == 0 {
-				return fmt.Errorf("SSH public key must be specified when running with AWS")
+				return fmt.Errorf("SSH public key must be specified when running with AWS (create with `kops create secret --name %s sshpublickey admin -i ~/.ssh/id_rsa.pub`)", cluster.Name)
+			}
+
+			if len(sshPublicKeys) != 1 {
+				return fmt.Errorf("Exactly one 'admin' SSH public key can be specified when running with AWS; please delete a key using `kops delete secret`")
+			} else {
+				l.Resources["ssh-public-key"] = fi.NewStringResource(string(sshPublicKeys[0]))
+
+				// SSHKeyName computes a unique SSH key name, combining the cluster name and the SSH public key fingerprint
+				l.TemplateFunctions["SSHKeyName"] = func() (string, error) {
+					fingerprint, err := awstasks.ComputeOpenSSHKeyFingerprint(string(sshPublicKeys[0]))
+					if err != nil {
+						return "", err
+					}
+
+					name := "kubernetes." + cluster.Name + "-" + fingerprint
+					return name, nil
+				}
 			}
 
 			l.TemplateFunctions["MachineTypeInfo"] = awsup.GetMachineTypeInfo
@@ -248,13 +273,18 @@ func (c *ApplyClusterCmd) Run() error {
 		return fmt.Errorf("unknown CloudProvider %q", cluster.Spec.CloudProvider)
 	}
 
+	clusterTags, err := buildClusterTags(cluster)
+	if err != nil {
+		return err
+	}
+
 	tf := &TemplateFunctions{
 		cluster: cluster,
-		tags:    tags,
+		tags:    clusterTags,
 		region:  region,
 	}
 
-	l.Tags = tags
+	l.Tags = clusterTags
 	l.WorkDir = c.OutDir
 	l.ModelStore = modelStore
 
@@ -265,26 +295,98 @@ func (c *ApplyClusterCmd) Run() error {
 		return secretStore
 	}
 
-	l.TemplateFunctions["ComputeNodeTags"] = func(args []string) []string {
+	// RenderNodeUpConfig returns the NodeUp config, in YAML format
-		var tags []string
+	l.TemplateFunctions["RenderNodeUpConfig"] = func(args []string) (string, error) {
-		for _, tag := range nodeUpTags {
+		var role api.InstanceGroupRole
-			tags = append(tags, tag)
+		for _, arg := range args {
+			if arg == "_kubernetes_master" {
+				if role != "" {
+					return "", fmt.Errorf("found duplicate role tags in args: %v", args)
+				}
+				role = api.InstanceGroupRoleMaster
+			}
+			if arg == "_kubernetes_pool" {
+				if role != "" {
+					return "", fmt.Errorf("found duplicate role tags in args: %v", args)
+				}
+				role = api.InstanceGroupRoleNode
+			}
+		}
+		if role == "" {
+			return "", fmt.Errorf("cannot determine role from args: %v", args)
 		}
 
-		isMaster := false
+		nodeUpTags, err := buildNodeupTags(role, tf.cluster, tf.tags)
-		for _, arg := range args {
+		if err != nil {
-			tags = append(tags, arg)
+			return "", err
-			if arg == "_kubernetes_master" {
+		}
-				isMaster = true
+
+		config := &nodeup.NodeUpConfig{}
+		for _, tag := range args {
+			config.Tags = append(config.Tags, tag)
+		}
+		for _, tag := range nodeUpTags {
+			config.Tags = append(config.Tags, tag)
+		}
+
+		config.Assets = c.Assets
+
+		config.ClusterName = cluster.Name
+
+		configPath, err := c.ClusterRegistry.ConfigurationPath(cluster.Name)
+		if err != nil {
+			return "", err
+		}
+		config.ClusterLocation = configPath.Path()
+
+		var images []*nodeup.Image
+
+		if isBaseURL(cluster.Spec.KubernetesVersion) {
+			baseURL := cluster.Spec.KubernetesVersion
+			baseURL = strings.TrimSuffix(baseURL, "/")
+
+			// TODO: pull kube-dns image
+			// When using a custom version, we want to preload the images over http
+			components := []string{"kube-proxy"}
+			if role == api.InstanceGroupRoleMaster {
+				components = append(components, "kube-apiserver", "kube-controller-manager", "kube-scheduler")
+			}
+			for _, component := range components {
+				imagePath := baseURL + "/bin/linux/amd64/" + component + ".tar"
+				glog.Infof("Adding docker image: %s", imagePath)
+
+				hash, err := findHash(imagePath)
+				if err != nil {
+					return "", err
+				}
+				image := &nodeup.Image{
+					Source: imagePath,
+					Hash:   hash.Hex(),
+				}
+				images = append(images, image)
 			}
 		}
 
-		if isMaster && !fi.BoolValue(cluster.Spec.IsolateMasters) {
+		config.Images = images
-			// Run this master as a pool node also (start kube-proxy etc)
+
-			tags = append(tags, "_kubernetes_pool")
+		{
+			protokubeImage := os.Getenv("PROTOKUBE_IMAGE")
+			if protokubeImage != "" {
+				glog.Warningf("Using protokube image specified in PROTOKUBE_IMAGE env var: %q", protokubeImage)
+			} else {
+				protokubeImage = nodeup.DefaultProtokubeImage
+			}
+			config.ProtokubeImage = &nodeup.Image{
+				Source: protokubeImage,
+			}
 		}
 
-		return tags
+		yaml, err := api.ToYaml(config)
+		if err != nil {
+			return "", err
+		}
+
+		return string(yaml), nil
 	}
 
 	//// TotalNodeCount computes the total count of nodes
@@ -335,28 +437,9 @@ func (c *ApplyClusterCmd) Run() error {
 	l.TemplateFunctions["NodeUpSourceHash"] = func() string {
 		return ""
 	}
-	l.TemplateFunctions["ClusterLocation"] = func() (string, error) {
-		configPath, err := c.ClusterRegistry.ConfigurationPath(cluster.Name)
-		if err != nil {
-			return "", err
-		}
-		return configPath.Path(), nil
-	}
-	l.TemplateFunctions["Assets"] = func() []string {
-		return c.Assets
-	}
 
 	tf.AddTo(l.TemplateFunctions)
 
-	if c.SSHPublicKey != "" {
-		authorized, err := ioutil.ReadFile(c.SSHPublicKey)
-		if err != nil {
-			return fmt.Errorf("error reading SSH key file %q: %v", c.SSHPublicKey, err)
-		}
-
-		l.Resources["ssh-public-key"] = fi.NewStringResource(string(authorized))
-	}
-
 	taskMap, err := l.BuildTasks(modelStore, c.Models)
 	if err != nil {
 		return fmt.Errorf("error building tasks: %v", err)
@@ -364,7 +447,7 @@ func (c *ApplyClusterCmd) Run() error {
 
 	var target fi.Target
 
-	switch c.Target {
+	switch c.TargetName {
 	case TargetDirect:
 		switch cluster.Spec.CloudProvider {
 		case "gce":
@@ -383,8 +466,9 @@ func (c *ApplyClusterCmd) Run() error {
 	case TargetDryRun:
 		target = fi.NewDryRunTarget(os.Stdout)
 	default:
-		return fmt.Errorf("unsupported target type %q", c.Target)
+		return fmt.Errorf("unsupported target type %q", c.TargetName)
 	}
+	c.Target = target
 
 	context, err := fi.NewContext(target, cloud, keyStore, secretStore, checkExisting)
 	if err != nil {
@@ -392,7 +476,7 @@ func (c *ApplyClusterCmd) Run() error {
 	}
 	defer context.Close()
 
-	err = context.RunTasks(taskMap)
+	err = context.RunTasks(taskMap, MaxAttemptsWithNoProgress)
 	if err != nil {
 		return fmt.Errorf("error running tasks: %v", err)
 	}
@@ -404,3 +488,23 @@ func (c *ApplyClusterCmd) Run() error {
 
 	return nil
 }
+
+func isBaseURL(kubernetesVersion string) bool {
+	return strings.HasPrefix(kubernetesVersion, "http:") || strings.HasPrefix(kubernetesVersion, "https:")
+}
+
+func findHash(url string) (*hashing.Hash, error) {
+	for _, ext := range []string{".sha1"} {
+		hashURL := url + ext
+		b, err := vfs.Context.ReadFile(hashURL)
+		if err != nil {
+			glog.Infof("error reading hash file %q: %v", hashURL, err)
+			continue
+		}
+		hashString := strings.TrimSpace(string(b))
+		glog.Infof("Found hash %q for %q", hashString, url)
+
+		return hashing.FromString(hashString)
+	}
+	return nil, fmt.Errorf("cannot determine hash for %v (have you specified a valid KubernetesVersion?)", url)
+}

upup/pkg/fi/cloudup/awstasks/instance.go

@@ -93,7 +93,7 @@ func (e *Instance) Find(c *fi.Context) (*Instance, error) {
 		request.Attribute = aws.String("userData")
 		response, err := cloud.EC2.DescribeInstanceAttribute(request)
 		if err != nil {
-			return nil, fmt.Errorf("error querying EC2 for user metadata for instance %q: %v", *i.InstanceId)
+			return nil, fmt.Errorf("error querying EC2 for user metadata for instance %q: %v", *i.InstanceId, err)
 		}
 		if response.UserData != nil {
 			b, err := base64.StdEncoding.DecodeString(aws.StringValue(response.UserData.Value))

upup/pkg/fi/cloudup/awstasks/sshkey.go

@@ -79,30 +79,28 @@ func (e *SSHKey) Find(c *fi.Context) (*SSHKey, error) {
 	return actual, nil
 }
 
-// computeAWSKeyFingerprint computes the AWS-specific fingerprint of the SSH public key
+// parseSSHPublicKey parses the SSH public key string
-func computeAWSKeyFingerprint(publicKey string) (string, error) {
+func parseSSHPublicKey(publicKey string) (ssh.PublicKey, error) {
 	tokens := strings.Fields(publicKey)
 	if len(tokens) < 2 {
-		return "", fmt.Errorf("error parsing SSH public key: %q", publicKey)
+		return nil, fmt.Errorf("error parsing SSH public key: %q", publicKey)
 	}
 
 	sshPublicKeyBytes, err := base64.StdEncoding.DecodeString(tokens[1])
 	if len(tokens) < 2 {
-		return "", fmt.Errorf("error decoding SSH public key: %q", publicKey)
+		return nil, fmt.Errorf("error decoding SSH public key: %q", publicKey)
 	}
 
 	sshPublicKey, err := ssh.ParsePublicKey(sshPublicKeyBytes)
 	if err != nil {
-		return "", fmt.Errorf("error parsing SSH public key: %v", err)
+		return nil, fmt.Errorf("error parsing SSH public key: %v", err)
 	}
+	return sshPublicKey, nil
+}
 
-	der, err := toDER(sshPublicKey)
+// colonSeparatedHex formats the byte slice SSH-fingerprint style: hex bytes separated by colons
-	if err != nil {
+func colonSeparatedHex(data []byte) string {
-		return "", fmt.Errorf("error computing fingerprint for SSH public key: %v", err)
+	sshKeyFingerprint := fmt.Sprintf("%x", data)
-	}
-	h := md5.Sum(der)
-	sshKeyFingerprint := fmt.Sprintf("%x", h)
-
 	var colonSeparated bytes.Buffer
 	for i := 0; i < len(sshKeyFingerprint); i++ {
 		if (i%2) == 0 && i != 0 {
@@ -111,7 +109,34 @@ func computeAWSKeyFingerprint(publicKey string) (string, error) {
 		colonSeparated.WriteByte(sshKeyFingerprint[i])
 	}
 
-	return colonSeparated.String(), nil
+	return colonSeparated.String()
+}
+
+// computeAWSKeyFingerprint computes the AWS-specific fingerprint of the SSH public key
+func computeAWSKeyFingerprint(publicKey string) (string, error) {
+	sshPublicKey, err := parseSSHPublicKey(publicKey)
+	if err != nil {
+		return "", err
+	}
+
+	der, err := toDER(sshPublicKey)
+	if err != nil {
+		return "", fmt.Errorf("error computing fingerprint for SSH public key: %v", err)
+	}
+	h := md5.Sum(der)
+
+	return colonSeparatedHex(h[:]), nil
+}
+
+// ComputeOpenSSHKeyFingerprint computes the OpenSSH fingerprint of the SSH public key
+func ComputeOpenSSHKeyFingerprint(publicKey string) (string, error) {
+	sshPublicKey, err := parseSSHPublicKey(publicKey)
+	if err != nil {
+		return "", err
+	}
+
+	h := md5.Sum(sshPublicKey.Marshal())
+	return colonSeparatedHex(h[:]), nil
 }
 
 // toDER gets the DER encoding of the SSH public key
@@ -195,6 +220,7 @@ func (_ *SSHKey) RenderAWS(t *awsup.AWSAPITarget, a, e, changes *SSHKey) error {
 		e.KeyFingerprint = response.KeyFingerprint
 	}
 
+	// No tags on SSH public key
 	return nil //return output.AddAWSTags(cloud.Tags(), v, "vpc")
 }
 
@@ -204,7 +230,8 @@ type terraformSSHKey struct {
 }
 
 func (_ *SSHKey) RenderTerraform(t *terraform.TerraformTarget, a, e, changes *SSHKey) error {
-	publicKey, err := t.AddFile("aws_key_pair", *e.Name, "public_key", e.PublicKey)
+	tfName := strings.Replace(*e.Name, ":", "", -1)
+	publicKey, err := t.AddFile("aws_key_pair", tfName, "public_key", e.PublicKey)
 	if err != nil {
 		return fmt.Errorf("error rendering PublicKey: %v", err)
 	}
@@ -214,9 +241,10 @@ func (_ *SSHKey) RenderTerraform(t *terraform.TerraformTarget, a, e, changes *SS
 		PublicKey: publicKey,
 	}
 
-	return t.RenderResource("aws_key_pair", *e.Name, tf)
+	return t.RenderResource("aws_key_pair", tfName, tf)
 }
 
 func (e *SSHKey) TerraformLink() *terraform.Literal {
-	return terraform.LiteralProperty("aws_key_pair", *e.Name, "id")
+	tfName := strings.Replace(*e.Name, ":", "", -1)
+	return terraform.LiteralProperty("aws_key_pair", tfName, "id")
 }

upup/pkg/fi/cloudup/awstasks/sshkey_test.go

@@ -6,7 +6,7 @@ import (
 	"testing"
 )
 
-func checkFingerprintEqual(t *testing.T, publicKey string, fingerprint string) {
+func checkAWSFingerprintEqual(t *testing.T, publicKey string, fingerprint string) {
 	actual, err := computeAWSKeyFingerprint(publicKey)
 	if err != nil {
 		t.Fatalf("Unexpected error computing AWS key fingerprint: %v", err)
@@ -16,7 +16,7 @@ func checkFingerprintEqual(t *testing.T, publicKey string, fingerprint string) {
 	}
 }
 
-func checkFingerprintError(t *testing.T, publicKey string, message string) {
+func checkAWSFingerprintError(t *testing.T, publicKey string, message string) {
 	_, err := computeAWSKeyFingerprint(publicKey)
 	if err == nil {
 		t.Fatalf("Expected error %q computing AWS key fingerprint", message)
@@ -27,29 +27,44 @@ func checkFingerprintError(t *testing.T, publicKey string, message string) {
 	}
 }
 
-func TestFingerprint_RsaKey1(t *testing.T) {
+func Test_AWSFingerprint_RsaKey1(t *testing.T) {
 	key := "ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCySdqIU+FhCWl3BNrAvPaOe5VfL2aCARUWwy91ZP+T7LBwFa9lhdttfjp/VX1D1/PVwntn2EhN079m8c2kfdmiZ/iCHqrLyIGSd+BOiCz0lT47znvANSfxYjLUuKrWWWeaXqerJkOsAD4PHchRLbZGPdbfoBKwtb/WT4GMRQmb9vmiaZYjsfdPPM9KkWI9ECoWFGjGehA8D+iYIPR711kRacb1xdYmnjHqxAZHFsb5L8wDWIeAyhy49cBD+lbzTiioq2xWLorXuFmXh6Do89PgzvHeyCLY6816f/kCX6wIFts8A2eaEHFL4rAOsuh6qHmSxGCR9peSyuRW8DxV725x justin@test"
-	checkFingerprintEqual(t, key, "85:a6:f4:64:b7:8f:4a:75:f1:ed:f9:26:1b:67:5f:f2")
+	checkAWSFingerprintEqual(t, key, "85:a6:f4:64:b7:8f:4a:75:f1:ed:f9:26:1b:67:5f:f2")
 }
 
-func TestFingerprint_RsaKeyEncrypted(t *testing.T) {
+func Test_AWSFingerprint_RsaKeyEncrypted(t *testing.T) {
 	// The private key is encrypted; the public key isn't
 	key := "ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCrLzpTNk5r3RWrzhRuFH8wkOQ+3mOEdaFosPzgDzQtriGU3JZ9Y3UHN4ltUOYUlapyFRaB27Pvyd48GkOSym7ZMn4/kyWn1SvXumJmW8bbX5+pTGK6p3Xu0elBPYMJHWEdZLK5gV6r15uRie9vhxknS9mOzxMcG9gdyyY3DdC3LiiRr6I8wTojP9MsWseZdPPZ5o6tMR/Zp2Q0fOb/DOhNuzunauMos+iu76YPORRFF1PaT1LoLxH7+/HwSX993JDzKytakuCoDFQ2/JvoMxkIvnVIz+MGsLKUZgmxJYQRaIL+fRR+ZBGFrOTqI72NXDmjT7aKjHHxYPfrsSggPh1J justin@machine"
-	checkFingerprintEqual(t, key, "c9:c5:05:5e:ea:54:fc:a4:7c:7c:75:5c:d2:71:5e:40")
+	checkAWSFingerprintEqual(t, key, "c9:c5:05:5e:ea:54:fc:a4:7c:7c:75:5c:d2:71:5e:40")
 }
 
-func TestFingerprint_TrickyWhitespace(t *testing.T) {
+func Test_AWSFingerprint_TrickyWhitespace(t *testing.T) {
 	// No name, \r instead of whitespace
 	key := "ssh-rsa\rAAAAB3NzaC1yc2EAAAADAQABAAABAQCySdqIU+FhCWl3BNrAvPaOe5VfL2aCARUWwy91ZP+T7LBwFa9lhdttfjp/VX1D1/PVwntn2EhN079m8c2kfdmiZ/iCHqrLyIGSd+BOiCz0lT47znvANSfxYjLUuKrWWWeaXqerJkOsAD4PHchRLbZGPdbfoBKwtb/WT4GMRQmb9vmiaZYjsfdPPM9KkWI9ECoWFGjGehA8D+iYIPR711kRacb1xdYmnjHqxAZHFsb5L8wDWIeAyhy49cBD+lbzTiioq2xWLorXuFmXh6Do89PgzvHeyCLY6816f/kCX6wIFts8A2eaEHFL4rAOsuh6qHmSxGCR9peSyuRW8DxV725x\r"
-	checkFingerprintEqual(t, key, "85:a6:f4:64:b7:8f:4a:75:f1:ed:f9:26:1b:67:5f:f2")
+	checkAWSFingerprintEqual(t, key, "85:a6:f4:64:b7:8f:4a:75:f1:ed:f9:26:1b:67:5f:f2")
 }
 
-func TestFingerprint_DsaKey(t *testing.T) {
+func Test_AWSFingerprint_DsaKey(t *testing.T) {
 	key := "ssh-dss AAAAB3NzaC1kc3MAAACBAIcCTu3vi9rNjsnhCrHeII7jSN6/FmnIdy09pQAsMAGGvCS9HBOteCKbIyYQQ0+Gi76Oui7cJ2VQojdxOxeZPoSP+QYnA+CVYhnowVVLeRA9VBQG3ZLInoXaqe3nR4/OXhY75GmYShBBPTQ+/fWGX9ltoXfygSc4KjhBNudvj75VAAAAFQDiw8A4MhY0aHSX/mtpa7XV8+iS6wAAAIAXyQaxM/dk0o1vBV3H0V0lGhog3mF7EJPdw7jagYvXQP1tAhzNofxZVhXHr4wGfiTQv9j5plDqQzCI/15a6DRyo9zI+zdPTR41W3dGrk56O2/Qxsz3/vNip5OwpOJ88yMmBX9m36gg0WrOXcZDgErhvZWRt5cXa9QjVg/KpxYLPAAAAIB8e5M82IiRLi+k1k4LsELKArQGzVkPgynESfnEXX0TKGiR7PJvBNGaKnPJtJ0Rrc38w/hLTeklroJt9Rdey/NI9b6tc+ur2pmJdnYppnNCm03WszU4oFD/7KIqR84Hf0fMbWd1hRvznpZhngZ505KNsL+ck0+Tlq6Hdhe2baXJcA== justin@machine"
-	checkFingerprintError(t, key, "AWS can only import RSA keys")
+	checkAWSFingerprintError(t, key, "AWS can only import RSA keys")
 }
 
-func TestFingerprint_Ed25519Key(t *testing.T) {
+func Test_AWSFingerprint_Ed25519Key(t *testing.T) {
 	key := "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFpyraYd4rUFftiEKzUO4wKFAgTkXxuJcRZwVcsuZJ8G justin@machine"
-	checkFingerprintError(t, key, "AWS can only import RSA keys")
+	checkAWSFingerprintError(t, key, "AWS can only import RSA keys")
+}
+
+func checkOpenSSHFingerprintEqual(t *testing.T, publicKey string, fingerprint string) {
+	actual, err := ComputeOpenSSHKeyFingerprint(publicKey)
+	if err != nil {
+		t.Fatalf("Unexpected error computing OpenSSH key fingerprint: %v", err)
+	}
+	if actual != fingerprint {
+		t.Fatalf("Expected fingerprint %q, got %q", fingerprint, actual)
+	}
+}
+
+func Test_OpenSSHFingerprint_RsaKey1(t *testing.T) {
+	key := "ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQCySdqIU+FhCWl3BNrAvPaOe5VfL2aCARUWwy91ZP+T7LBwFa9lhdttfjp/VX1D1/PVwntn2EhN079m8c2kfdmiZ/iCHqrLyIGSd+BOiCz0lT47znvANSfxYjLUuKrWWWeaXqerJkOsAD4PHchRLbZGPdbfoBKwtb/WT4GMRQmb9vmiaZYjsfdPPM9KkWI9ECoWFGjGehA8D+iYIPR711kRacb1xdYmnjHqxAZHFsb5L8wDWIeAyhy49cBD+lbzTiioq2xWLorXuFmXh6Do89PgzvHeyCLY6816f/kCX6wIFts8A2eaEHFL4rAOsuh6qHmSxGCR9peSyuRW8DxV725x justin@test"
+	checkOpenSSHFingerprintEqual(t, key, "be:ba:ec:2b:9e:a0:68:b8:19:6b:9a:26:cc:b1:58:ff")
 }

upup/pkg/fi/cloudup/awsup/aws_cloud.go

@@ -78,7 +78,7 @@ func (c *AWSCloud) Tags() map[string]string {
 func isTagsEventualConsistencyError(err error) bool {
 	if awsErr, ok := err.(awserr.Error); ok {
 		switch awsErr.Code() {
-		case "InvalidInstanceID.NotFound", "InvalidRouteTableID.NotFound", "InvalidVpcID.NotFound", "InvalidGroup.NotFound", "InvalidSubnetID.NotFound", "InvalidInternetGatewayID.NotFound":
+		case "InvalidInstanceID.NotFound", "InvalidRouteTableID.NotFound", "InvalidVpcID.NotFound", "InvalidGroup.NotFound", "InvalidSubnetID.NotFound", "InvalidInternetGatewayID.NotFound", "InvalidDhcpOptionsID.NotFound":
 			return true
 
 		default:
@@ -520,5 +520,5 @@ func (c *AWSCloud) FindDNSHostedZone(clusterDNSName string) (string, error) {
 		return id, nil
 	}
 
-	return "", fmt.Errorf("Found multiple hosted zones matching cluster %q; please specify the ID of the zone to use")
+	return "", fmt.Errorf("Found multiple hosted zones matching cluster %q; please specify the ID of the zone to use", clusterDNSName)
 }

upup/pkg/fi/cloudup/awsup/machine_types.go

@@ -239,6 +239,71 @@ var MachineTypes []AWSMachineTypeInfo = []AWSMachineTypeInfo{
 		EphemeralDisks: nil,
 	},
 
+	// cc2 family
+	{
+		Name:           "cc2.8xlarge",
+		MemoryGB:       60.5,
+		ECU:            88,
+		Cores:          32,
+		EphemeralDisks: []int{840, 840, 840, 840},
+	},
+
+	// cg1 family
+	{
+		Name:           "cg1.4xlarge",
+		MemoryGB:       22.5,
+		ECU:            33.5,
+		Cores:          16,
+		EphemeralDisks: []int{840, 840},
+	},
+
+	// cr1 family
+	{
+		Name:           "cr1.8xlarge",
+		MemoryGB:       244.0,
+		ECU:            88,
+		Cores:          32,
+		EphemeralDisks: []int{120, 120},
+	},
+
+	// d2 family
+	{
+		Name:           "d2.xlarge",
+		MemoryGB:       30.5,
+		ECU:            14,
+		Cores:          4,
+		EphemeralDisks: []int{2000, 2000, 2000},
+	},
+	{
+		Name:           "d2.2xlarge",
+		MemoryGB:       61.0,
+		ECU:            28,
+		Cores:          8,
+		EphemeralDisks: []int{2000, 2000, 2000, 2000, 2000, 2000},
+	},
+	{
+		Name:     "d2.4xlarge",
+		MemoryGB: 122.0,
+		ECU:      56,
+		Cores:    16,
+		EphemeralDisks: []int{
+			2000, 2000, 2000, 2000, 2000, 2000,
+			2000, 2000, 2000, 2000, 2000, 2000,
+		},
+	},
+	{
+		Name:     "d2.8xlarge",
+		MemoryGB: 244.0,
+		ECU:      116,
+		Cores:    36,
+		EphemeralDisks: []int{
+			2000, 2000, 2000, 2000, 2000, 2000,
+			2000, 2000, 2000, 2000, 2000, 2000,
+			2000, 2000, 2000, 2000, 2000, 2000,
+			2000, 2000, 2000, 2000, 2000, 2000,
+		},
+	},
+
 	// g2 family
 	{
 		Name:           "g2.2xlarge",
@@ -255,6 +320,15 @@ var MachineTypes []AWSMachineTypeInfo = []AWSMachineTypeInfo{
 		EphemeralDisks: []int{120, 120},
 	},
 
+	// hi1 family
+	{
+		Name:           "hi1.4xlarge",
+		MemoryGB:       60.5,
+		ECU:            35,
+		Cores:          16,
+		EphemeralDisks: []int{1024, 1024},
+	},
+
 	// i2 family
 	{
 		Name:           "i2.xlarge",

upup/pkg/fi/cloudup/networking.go

@@ -0,0 +1,28 @@
+package cloudup
+
+import (
+	"github.com/golang/glog"
+	"k8s.io/kops/upup/pkg/api"
+)
+
+func usesCNI(c *api.Cluster) bool {
+	networkConfig := c.Spec.Networking
+	if networkConfig == nil || networkConfig.Classic != nil {
+		// classic
+		return false
+	}
+
+	if networkConfig.Kubenet != nil {
+		// kubenet
+		return true
+	}
+
+	if networkConfig.External != nil {
+		// external: assume uses CNI
+		return true
+	}
+
+	// Assume other modes also use CNI
+	glog.Warningf("Unknown networking mode configured")
+	return true
+}

upup/pkg/fi/cloudup/populate_cluster_spec.go

@@ -6,7 +6,6 @@ import (
 	"github.com/golang/glog"
 	"k8s.io/kops/upup/pkg/api"
 	"k8s.io/kops/upup/pkg/fi"
-	"k8s.io/kops/upup/pkg/fi/hashing"
 	"k8s.io/kops/upup/pkg/fi/loader"
 	"k8s.io/kops/upup/pkg/fi/utils"
 	"k8s.io/kops/upup/pkg/fi/vfs"
@@ -87,10 +86,6 @@ func PopulateClusterSpec(cluster *api.Cluster, clusterRegistry *api.ClusterRegis
 }
 
 func (c *populateClusterSpec) run() error {
-	// TODO: Make these configurable?
-	useMasterASG := true
-	useMasterLB := false
-
 	err := c.InputCluster.Validate(false)
 	if err != nil {
 		return err
@@ -170,8 +165,6 @@ func (c *populateClusterSpec) run() error {
 		return fmt.Errorf("ClusterRegistry is required")
 	}
 
-	tags := make(map[string]struct{})
-
 	keyStore := c.ClusterRegistry.KeyStore(cluster.Name)
 	// Always assume a dry run during this phase
 	keyStore.(*fi.VFSCAStore).DryRun = true
@@ -223,19 +216,6 @@ func (c *populateClusterSpec) run() error {
 		// We do support this...
 	}
 
-	if cluster.Spec.KubernetesVersion == "" {
-		stableURL := "https://storage.googleapis.com/kubernetes-release/release/stable.txt"
-		b, err := vfs.Context.ReadFile(stableURL)
-		if err != nil {
-			return fmt.Errorf("--kubernetes-version not specified, and unable to download latest version from %q: %v", stableURL, err)
-		}
-		latestVersion := strings.TrimSpace(string(b))
-		glog.Infof("Using kubernetes latest stable version: %s", latestVersion)
-
-		cluster.Spec.KubernetesVersion = latestVersion
-		//return fmt.Errorf("Must either specify a KubernetesVersion (-kubernetes-version) or provide an asset with the release bundle")
-	}
-
 	// Normalize k8s version
 	versionWithoutV := strings.TrimSpace(cluster.Spec.KubernetesVersion)
 	if strings.HasPrefix(versionWithoutV, "v") {
@@ -246,47 +226,11 @@ func (c *populateClusterSpec) run() error {
 		cluster.Spec.KubernetesVersion = versionWithoutV
 	}
 
-	if useMasterASG {
-		tags["_master_asg"] = struct{}{}
-	} else {
-		tags["_master_single"] = struct{}{}
-	}
-
-	if useMasterLB {
-		tags["_master_lb"] = struct{}{}
-	} else {
-		tags["_not_master_lb"] = struct{}{}
-	}
-
-	if cluster.Spec.MasterPublicName != "" {
-		tags["_master_dns"] = struct{}{}
-	}
-
-	if fi.BoolValue(cluster.Spec.IsolateMasters) {
-		tags["_isolate_masters"] = struct{}{}
-	}
-
 	cloud, err := BuildCloud(cluster)
 	if err != nil {
 		return err
 	}
 
-	switch cluster.Spec.CloudProvider {
-	case "gce":
-		{
-			glog.Fatalf("GCE is (probably) not working currently - please ping @justinsb for cleanup")
-			tags["_gce"] = struct{}{}
-		}
-
-	case "aws":
-		{
-			tags["_aws"] = struct{}{}
-		}
-
-	default:
-		return fmt.Errorf("unknown CloudProvider %q", cluster.Spec.CloudProvider)
-	}
-
 	if cluster.Spec.DNSZone == "" {
 		dnsZone, err := cloud.FindDNSHostedZone(cluster.Name)
 		if err != nil {
@@ -296,6 +240,11 @@ func (c *populateClusterSpec) run() error {
 		cluster.Spec.DNSZone = dnsZone
 	}
 
+	tags, err := buildClusterTags(cluster)
+	if err != nil {
+		return err
+	}
+
 	tf := &TemplateFunctions{
 		cluster: cluster,
 		tags:    tags,
@@ -329,22 +278,6 @@ func (c *populateClusterSpec) run() error {
 	return nil
 }
 
-func findHash(url string) (*hashing.Hash, error) {
-	for _, ext := range []string{".sha1"} {
-		hashURL := url + ext
-		b, err := vfs.Context.ReadFile(hashURL)
-		if err != nil {
-			glog.Infof("error reading hash file %q: %v", hashURL, err)
-			continue
-		}
-		hashString := strings.TrimSpace(string(b))
-		glog.Infof("Found hash %q for %q", hashString, url)
-
-		return hashing.FromString(hashString)
-	}
-	return nil, fmt.Errorf("cannot determine hash for %v (have you specified a valid KubernetesVersion?)", url)
-}
-
 func (c *populateClusterSpec) assignSubnets(cluster *api.Cluster) error {
 	if cluster.Spec.NonMasqueradeCIDR == "" {
 		glog.Warningf("NonMasqueradeCIDR not set; can't auto-assign dependent subnets")

upup/pkg/fi/cloudup/populatecluster_test.go

@@ -28,6 +28,26 @@ func buildMinimalCluster() *api.Cluster {
 	return c
 }
 
+func addEtcdClusters(c *api.Cluster) {
+	zones := sets.NewString()
+	for _, z := range c.Spec.Zones {
+		zones.Insert(z.Name)
+	}
+	etcdZones := zones.List()
+
+	for _, etcdCluster := range EtcdClusters {
+		etcd := &api.EtcdClusterSpec{}
+		etcd.Name = etcdCluster
+		for _, zone := range etcdZones {
+			m := &api.EtcdMemberSpec{}
+			m.Name = zone
+			m.Zone = zone
+			etcd.Members = append(etcd.Members, m)
+		}
+		c.Spec.EtcdClusters = append(c.Spec.EtcdClusters, etcd)
+	}
+}
+
 func TestPopulateCluster_Default_NoError(t *testing.T) {
 	c := buildMinimalCluster()
 
@@ -36,25 +56,7 @@ func TestPopulateCluster_Default_NoError(t *testing.T) {
 		t.Fatalf("error from PerformAssignments: %v", err)
 	}
 
-	if len(c.Spec.EtcdClusters) == 0 {
+	addEtcdClusters(c)
-		zones := sets.NewString()
-		for _, z := range c.Spec.Zones {
-			zones.Insert(z.Name)
-		}
-		etcdZones := zones.List()
-
-		for _, etcdCluster := range EtcdClusters {
-			etcd := &api.EtcdClusterSpec{}
-			etcd.Name = etcdCluster
-			for _, zone := range etcdZones {
-				m := &api.EtcdMemberSpec{}
-				m.Name = zone
-				m.Zone = zone
-				etcd.Members = append(etcd.Members, m)
-			}
-			c.Spec.EtcdClusters = append(c.Spec.EtcdClusters, etcd)
-		}
-	}
 
 	registry := buildInmemoryClusterRegistry()
 	_, err = PopulateClusterSpec(c, registry)
@@ -63,6 +65,76 @@ func TestPopulateCluster_Default_NoError(t *testing.T) {
 	}
 }
 
+func TestPopulateCluster_Docker_Spec(t *testing.T) {
+	c := buildMinimalCluster()
+	c.Spec.Docker = &api.DockerConfig{
+		MTU:              5678,
+		InsecureRegistry: "myregistry.com:1234",
+	}
+
+	err := c.PerformAssignments()
+	if err != nil {
+		t.Fatalf("error from PerformAssignments: %v", err)
+	}
+
+	addEtcdClusters(c)
+
+	registry := buildInmemoryClusterRegistry()
+	full, err := PopulateClusterSpec(c, registry)
+	if err != nil {
+		t.Fatalf("Unexpected error from PopulateCluster: %v", err)
+	}
+
+	if full.Spec.Docker.MTU != 5678 {
+		t.Fatalf("Unexpected Docker MTU: %v", full.Spec.Docker.MTU)
+	}
+
+	if full.Spec.Docker.InsecureRegistry != "myregistry.com:1234" {
+		t.Fatalf("Unexpected Docker InsecureRegistry: %v", full.Spec.Docker.InsecureRegistry)
+	}
+
+	// Check default values not changed
+	if full.Spec.Docker.Bridge != "cbr0" {
+		t.Fatalf("Unexpected Docker Bridge: %v", full.Spec.Docker.Bridge)
+	}
+}
+
+func build(c *api.Cluster) (*api.Cluster, error) {
+	err := c.PerformAssignments()
+	if err != nil {
+		return nil, fmt.Errorf("error from PerformAssignments: %v", err)
+	}
+
+	addEtcdClusters(c)
+	registry := buildInmemoryClusterRegistry()
+	full, err := PopulateClusterSpec(c, registry)
+	if err != nil {
+		return nil, fmt.Errorf("Unexpected error from PopulateCluster: %v", err)
+	}
+	return full, nil
+}
+
+func TestPopulateCluster_Kubenet(t *testing.T) {
+	c := buildMinimalCluster()
+
+	full, err := build(c)
+	if err != nil {
+		t.Fatal("error during build: %v", err)
+	}
+
+	if full.Spec.Kubelet.NetworkPluginName != "kubenet" {
+		t.Fatalf("Unexpected NetworkPluginName: %v", full.Spec.Kubelet.NetworkPluginName)
+	}
+
+	if fi.BoolValue(full.Spec.Kubelet.ReconcileCIDR) != true {
+		t.Fatalf("Unexpected ReconcileCIDR: %v", full.Spec.Kubelet.ReconcileCIDR)
+	}
+
+	if fi.BoolValue(full.Spec.KubeControllerManager.ConfigureCloudRoutes) != true {
+		t.Fatalf("Unexpected ConfigureCloudRoutes: %v", full.Spec.KubeControllerManager.ConfigureCloudRoutes)
+	}
+}
+
 func TestPopulateCluster_Custom_CIDR(t *testing.T) {
 	c := buildMinimalCluster()
 	c.Spec.NetworkCIDR = "172.20.2.0/24"
@@ -77,25 +149,7 @@ func TestPopulateCluster_Custom_CIDR(t *testing.T) {
 		t.Fatalf("error from PerformAssignments: %v", err)
 	}
 
-	if len(c.Spec.EtcdClusters) == 0 {
+	addEtcdClusters(c)
-		zones := sets.NewString()
-		for _, z := range c.Spec.Zones {
-			zones.Insert(z.Name)
-		}
-		etcdZones := zones.List()
-
-		for _, etcdCluster := range EtcdClusters {
-			etcd := &api.EtcdClusterSpec{}
-			etcd.Name = etcdCluster
-			for _, zone := range etcdZones {
-				m := &api.EtcdMemberSpec{}
-				m.Name = zone
-				m.Zone = zone
-				etcd.Members = append(etcd.Members, m)
-			}
-			c.Spec.EtcdClusters = append(c.Spec.EtcdClusters, etcd)
-		}
-	}
 
 	registry := buildInmemoryClusterRegistry()
 	full, err := PopulateClusterSpec(c, registry)
@@ -116,25 +170,7 @@ func TestPopulateCluster_IsolateMasters(t *testing.T) {
 		t.Fatalf("error from PerformAssignments: %v", err)
 	}
 
-	if len(c.Spec.EtcdClusters) == 0 {
+	addEtcdClusters(c)
-		zones := sets.NewString()
-		for _, z := range c.Spec.Zones {
-			zones.Insert(z.Name)
-		}
-		etcdZones := zones.List()
-
-		for _, etcdCluster := range EtcdClusters {
-			etcd := &api.EtcdClusterSpec{}
-			etcd.Name = etcdCluster
-			for _, zone := range etcdZones {
-				m := &api.EtcdMemberSpec{}
-				m.Name = zone
-				m.Zone = zone
-				etcd.Members = append(etcd.Members, m)
-			}
-			c.Spec.EtcdClusters = append(c.Spec.EtcdClusters, etcd)
-		}
-	}
 
 	registry := buildInmemoryClusterRegistry()
 	full, err := PopulateClusterSpec(c, registry)
@@ -158,25 +194,7 @@ func TestPopulateCluster_IsolateMastersFalse(t *testing.T) {
 		t.Fatalf("error from PerformAssignments: %v", err)
 	}
 
-	if len(c.Spec.EtcdClusters) == 0 {
+	addEtcdClusters(c)
-		zones := sets.NewString()
-		for _, z := range c.Spec.Zones {
-			zones.Insert(z.Name)
-		}
-		etcdZones := zones.List()
-
-		for _, etcdCluster := range EtcdClusters {
-			etcd := &api.EtcdClusterSpec{}
-			etcd.Name = etcdCluster
-			for _, zone := range etcdZones {
-				m := &api.EtcdMemberSpec{}
-				m.Name = zone
-				m.Zone = zone
-				etcd.Members = append(etcd.Members, m)
-			}
-			c.Spec.EtcdClusters = append(c.Spec.EtcdClusters, etcd)
-		}
-	}
 
 	registry := buildInmemoryClusterRegistry()
 	full, err := PopulateClusterSpec(c, registry)

upup/pkg/fi/cloudup/tagbuilder.go

@@ -0,0 +1,106 @@
+package cloudup
+
+import (
+	"fmt"
+	"github.com/golang/glog"
+	"k8s.io/kops/upup/pkg/api"
+	"k8s.io/kops/upup/pkg/fi"
+)
+
+func buildClusterTags(cluster *api.Cluster) (map[string]struct{}, error) {
+	// TODO: Make these configurable?
+	useMasterASG := true
+	useMasterLB := false
+
+	tags := make(map[string]struct{})
+
+	networking := cluster.Spec.Networking
+	if networking == nil || networking.Classic != nil {
+		tags["_networking_classic"] = struct{}{}
+	} else if networking.Kubenet != nil {
+		tags["_networking_kubenet"] = struct{}{}
+	} else if networking.External != nil {
+		// external is based on kubenet
+		tags["_networking_kubenet"] = struct{}{}
+		tags["_networking_external"] = struct{}{}
+	} else {
+		return nil, fmt.Errorf("No networking mode set")
+	}
+
+	if useMasterASG {
+		tags["_master_asg"] = struct{}{}
+	} else {
+		tags["_master_single"] = struct{}{}
+	}
+
+	if useMasterLB {
+		tags["_master_lb"] = struct{}{}
+	} else {
+		tags["_not_master_lb"] = struct{}{}
+	}
+
+	if cluster.Spec.MasterPublicName != "" {
+		tags["_master_dns"] = struct{}{}
+	}
+
+	if fi.BoolValue(cluster.Spec.IsolateMasters) {
+		tags["_isolate_masters"] = struct{}{}
+	}
+
+	switch cluster.Spec.CloudProvider {
+	case "gce":
+		{
+			glog.Fatalf("GCE is (probably) not working currently - please ping @justinsb for cleanup")
+			tags["_gce"] = struct{}{}
+		}
+
+	case "aws":
+		{
+			tags["_aws"] = struct{}{}
+		}
+
+	default:
+		return nil, fmt.Errorf("unknown CloudProvider %q", cluster.Spec.CloudProvider)
+	}
+
+	return tags, nil
+}
+
+func buildNodeupTags(role api.InstanceGroupRole, cluster *api.Cluster, clusterTags map[string]struct{}) ([]string, error) {
+	var tags []string
+
+	switch role {
+	case api.InstanceGroupRoleNode:
+		// No special tags
+
+		// TODO: Should we run _protokube on the nodes?
+		tags = append(tags, "_protokube")
+
+	case api.InstanceGroupRoleMaster:
+		if !fi.BoolValue(cluster.Spec.IsolateMasters) {
+			// Run this master as a pool node also (start kube-proxy etc)
+			tags = append(tags, "_kubernetes_pool")
+		}
+		tags = append(tags, "_protokube")
+	default:
+		return nil, fmt.Errorf("Unrecognized role: %v", role)
+	}
+
+	// TODO: Replace with list of CNI plugins ?
+	if usesCNI(cluster) {
+		tags = append(tags, "_cni_bridge")
+		tags = append(tags, "_cni_host_local")
+		tags = append(tags, "_cni_loopback")
+		tags = append(tags, "_cni_ptp")
+		//tags = append(tags, "_cni_tuning")
+	}
+
+	if _, found := clusterTags["_gce"]; found {
+		tags = append(tags, "_gce")
+	}
+	if _, found := clusterTags["_aws"]; found {
+		tags = append(tags, "_aws")
+	}
+
+	return tags, nil
+}

upup/pkg/fi/cloudup/tagbuilder_test.go

@@ -0,0 +1,41 @@
+package cloudup
+
+import (
+	"k8s.io/kops/upup/pkg/api"
+	"testing"
+)
+
+func TestBuildTags_CloudProvider_AWS(t *testing.T) {
+	c := &api.Cluster{
+		Spec: api.ClusterSpec{
+			CloudProvider: "aws",
+		},
+	}
+
+	tags, err := buildClusterTags(c)
+	if err != nil {
+		t.Fatalf("buildTags error: %v", err)
+	}
+
+	if _, found := tags["_aws"]; !found {
+		t.Fatalf("tag _aws not found")
+	}
+
+	nodeUpTags, err := buildNodeupTags(api.InstanceGroupRoleNode, c, tags)
+	if err != nil {
+		t.Fatalf("buildNodeupTags error: %v", err)
+	}
+
+	if !stringSliceContains(nodeUpTags, "_aws") {
+		t.Fatalf("nodeUpTag _aws not found")
+	}
+}
+
+func stringSliceContains(haystack []string, needle string) bool {
+	for _, s := range haystack {
+		if needle == s {
+			return true
+		}
+	}
+	return false
+}

upup/pkg/fi/cloudup/template_functions.go

@@ -4,7 +4,9 @@ import (
 	"encoding/base64"
 	"encoding/binary"
 	"fmt"
+	"github.com/golang/glog"
 	"k8s.io/kops/upup/pkg/api"
+	"k8s.io/kops/upup/pkg/fi/vfs"
 	"math/big"
 	"net"
 	"sort"
@@ -78,6 +80,7 @@ func (tf *TemplateFunctions) AddTo(dest template.FuncMap) {
 	dest["IAMPrefix"] = tf.IAMPrefix
 	dest["IAMServiceEC2"] = tf.IAMServiceEC2
 
+	dest["Image"] = tf.Image
 }
 
 func (tf *TemplateFunctions) EtcdClusterMemberTags(etcd *api.EtcdClusterSpec, m *api.EtcdMemberSpec) map[string]string {
@@ -137,3 +140,30 @@ func (tf *TemplateFunctions) IAMPrefix() string {
 		return "arn:aws"
 	}
 }
+
+// Image returns the docker image name for the specified component
+func (tf *TemplateFunctions) Image(component string) (string, error) {
+	if component == "kube-dns" {
+		// TODO: Once we are shipping different versions, start to use them
+		return "gcr.io/google_containers/kubedns-amd64:1.3", nil
+	}
+
+	if !isBaseURL(tf.cluster.Spec.KubernetesVersion) {
+		return "gcr.io/google_containers/" + component + ":" + "v" + tf.cluster.Spec.KubernetesVersion, nil
+	}
+
+	baseURL := tf.cluster.Spec.KubernetesVersion
+	baseURL = strings.TrimSuffix(baseURL, "/")
+
+	tagURL := baseURL + "/bin/linux/amd64/" + component + ".docker_tag"
+	glog.V(2).Infof("Downloading docker tag for %s from: %s", component, tagURL)
+
+	b, err := vfs.Context.ReadFile(tagURL)
+	if err != nil {
+		return "", fmt.Errorf("error reading tag file %q: %v", tagURL, err)
+	}
+	tag := strings.TrimSpace(string(b))
+	glog.V(2).Infof("Found tag %q for %q", tag, component)
+
+	return "gcr.io/google_containers/" + component + ":" + tag, nil
+}

upup/pkg/fi/cloudup/validation_test.go

@@ -104,6 +104,18 @@ func TestValidateFull_Default_Validates(t *testing.T) {
 	}
 }
 
+func TestValidateFull_ClusterName_InvalidDNS_NoDot(t *testing.T) {
+	c := buildDefaultCluster(t)
+	c.Name = "test"
+	expectErrorFromValidate(t, c, "DNS name")
+}
+
+func TestValidateFull_ClusterName_InvalidDNS_Invalid(t *testing.T) {
+	c := buildDefaultCluster(t)
+	c.Name = "test.-"
+	expectErrorFromValidate(t, c, "DNS name")
+}
+
 func TestValidateFull_ClusterName_Required(t *testing.T) {
 	c := buildDefaultCluster(t)
 	c.Name = ""

upup/pkg/fi/context.go

@@ -38,11 +38,11 @@ func NewContext(target Target, cloud Cloud, castore CAStore, secretStore SecretS
 	return c, nil
 }
 
-func (c *Context) RunTasks(taskMap map[string]Task) error {
+func (c *Context) RunTasks(taskMap map[string]Task, maxAttemptsWithNoProgress int) error {
 	e := &executor{
 		context: c,
 	}
-	return e.RunTasks(taskMap)
+	return e.RunTasks(taskMap, maxAttemptsWithNoProgress)
 }
 
 func (c *Context) Close() {

upup/pkg/fi/dryrun_target.go

@@ -257,3 +257,8 @@ func ValueAsString(value reflect.Value) string {
 func (t *DryRunTarget) Finish(taskMap map[string]Task) error {
 	return t.PrintReport(taskMap, t.out)
 }
+
+// HasChanges returns true iff any changes would have been made
+func (t *DryRunTarget) HasChanges() bool {
+	return len(t.changes) != 0
+}

upup/pkg/fi/executor.go

@@ -8,8 +8,6 @@ import (
 	"time"
 )
 
-const MaxAttemptsWithNoProgress = 3
-
 type executor struct {
 	context *Context
 }
@@ -22,8 +20,8 @@ type taskState struct {
 }
 
 // RunTasks executes all the tasks, considering their dependencies
-// It will perform some re-execution on error, retrying as long as progess is still being made
+// It will perform some re-execution on error, retrying as long as progress is still being made
-func (e *executor) RunTasks(taskMap map[string]Task) error {
+func (e *executor) RunTasks(taskMap map[string]Task, maxAttemptsWithNoProgress int) error {
 	dependencies := FindTaskDependencies(taskMap)
 
 	taskStates := make(map[string]*taskState)
@@ -94,7 +92,7 @@ func (e *executor) RunTasks(taskMap map[string]Task) error {
 		if !progress {
 			if len(errors) != 0 {
 				noProgressCount++
-				if noProgressCount == MaxAttemptsWithNoProgress {
+				if noProgressCount == maxAttemptsWithNoProgress {
 					return fmt.Errorf("did not make any progress executing task.  Example error: %v", errors[0])
 				} else {
 					glog.Infof("No progress made, sleeping before retrying failed tasks")

upup/pkg/fi/hashing/hash.go

@@ -65,7 +65,7 @@ func (ha HashAlgorithm) FromString(s string) (*Hash, error) {
 	}
 
 	if len(s) != l {
-		return nil, fmt.Errorf("invalid %q hash - unexpected length %s", ha, len(s))
+		return nil, fmt.Errorf("invalid %q hash - unexpected length %d", ha, len(s))
 	}
 
 	hashValue, err := hex.DecodeString(s)

upup/pkg/fi/nodeup/build_flags.go

@@ -15,12 +15,13 @@ func buildFlags(options interface{}) (string, error) {
 
 	walker := func(path string, field *reflect.StructField, val reflect.Value) error {
 		if field == nil {
-			glog.V(4).Infof("not writing non-field: %s", path)
+			glog.V(8).Infof("ignoring non-field: %s", path)
 			return nil
 		}
 		tag := field.Tag.Get("flag")
 		if tag == "" {
 			glog.V(4).Infof("not writing field with no flag tag: %s", path)
+			// We want to descend - it could be a structure containing flags
 			return nil
 		}
 		if tag == "-" {
@@ -29,6 +30,8 @@ func buildFlags(options interface{}) (string, error) {
 		}
 		flagName := tag
 
+		// We do have to do this, even though the recursive walk will do it for us
+		// because when we descend we won't have `field` set
 		if val.Kind() == reflect.Ptr {
 			if val.IsNil() {
 				return nil
@@ -54,7 +57,8 @@ func buildFlags(options interface{}) (string, error) {
 		if flag != "" {
 			flags = append(flags, flag)
 		}
-		return nil
+		// Nothing more to do here
+		return utils.SkipReflection
 	}
 	err := utils.ReflectRecursive(reflect.ValueOf(options), walker)
 	if err != nil {

upup/pkg/fi/nodeup/command.go

@@ -8,17 +8,22 @@ import (
 	"k8s.io/kops/upup/pkg/fi"
 	"k8s.io/kops/upup/pkg/fi/nodeup/cloudinit"
 	"k8s.io/kops/upup/pkg/fi/nodeup/local"
+	"k8s.io/kops/upup/pkg/fi/nodeup/nodetasks"
 	"k8s.io/kops/upup/pkg/fi/utils"
 	"k8s.io/kops/upup/pkg/fi/vfs"
+	"strconv"
 	"strings"
 )
 
+// We should probably retry for a long time - there is not really any great fallback
+const MaxAttemptsWithNoProgress = 100
+
 type NodeUpCommand struct {
 	config         *NodeUpConfig
 	cluster        *api.Cluster
 	ConfigLocation string
 	ModelDir       string
-	AssetDir       string
+	CacheDir       string
 	Target         string
 	FSRoot         string
 }
@@ -42,10 +47,10 @@ func (c *NodeUpCommand) Run(out io.Writer) error {
 		return fmt.Errorf("ConfigLocation is required")
 	}
 
-	if c.AssetDir == "" {
+	if c.CacheDir == "" {
-		return fmt.Errorf("AssetDir is required")
+		return fmt.Errorf("CacheDir is required")
 	}
-	assets := fi.NewAssetStore(c.AssetDir)
+	assets := fi.NewAssetStore(c.CacheDir)
 	for _, asset := range c.config.Assets {
 		err := assets.Add(asset)
 		if err != nil {
@@ -135,6 +140,13 @@ func (c *NodeUpCommand) Run(out io.Writer) error {
 		return fmt.Errorf("error building loader: %v", err)
 	}
 
+	for i, image := range c.config.Images {
+		taskMap["LoadImage."+strconv.Itoa(i)] = &nodetasks.LoadImageTask{
+			Source: image.Source,
+			Hash:   image.Hash,
+		}
+	}
+
 	var cloud fi.Cloud
 	var caStore fi.CAStore
 	var secretStore fi.SecretStore
@@ -143,7 +155,9 @@ func (c *NodeUpCommand) Run(out io.Writer) error {
 
 	switch c.Target {
 	case "direct":
-		target = &local.LocalTarget{}
+		target = &local.LocalTarget{
+			CacheDir: c.CacheDir,
+		}
 	case "dryrun":
 		target = fi.NewDryRunTarget(out)
 	case "cloudinit":
@@ -159,7 +173,7 @@ func (c *NodeUpCommand) Run(out io.Writer) error {
 	}
 	defer context.Close()
 
-	err = context.RunTasks(taskMap)
+	err = context.RunTasks(taskMap, MaxAttemptsWithNoProgress)
 	if err != nil {
 		glog.Exitf("error running tasks: %v", err)
 	}

upup/pkg/fi/nodeup/config.go

@@ -11,6 +11,9 @@ type NodeUpConfig struct {
 	// TODO: Remove once everything is in containers?
 	Assets []string `json:",omitempty"`
 
+	// Images are a list of images we should preload
+	Images []*Image `json:"images,omitempty"`
+
 	// ClusterLocation is the VFS path to the cluster spec
 	ClusterLocation string `json:",omitempty"`
 
@@ -18,40 +21,16 @@ type NodeUpConfig struct {
 	// Technically this is redundant - it is in ClusterLocation, but this can serve as a cross-check,
 	// and it allows us to more easily identify the cluster, for example when we are deleting resources.
 	ClusterName string `json:",omitempty"`
+
+	// ProtokubeImage is the docker image to load for protokube (bootstrapping)
+	ProtokubeImage *Image `json:"protokubeImage"`
 }
 
-// Our client configuration structure
+// Image is a docker image we should pre-load
-// Wherever possible, we try to use the types & names in https://github.com/kubernetes/kubernetes/blob/master/pkg/apis/componentconfig/types.go
+type Image struct {
+	// Source is the URL from which we should download the image
+	Source string `json:"source,omitempty"`
 
-//type NodeConfig struct {
+	// Hash is the hash of the file, to verify image integrity (even over http)
-//	Kubelet               *KubeletConfig               `json:",omitempty"`
+	Hash string `json:"hash,omitempty"`
-//	KubeProxy             *KubeProxyConfig             `json:",omitempty"`
+}
-//	KubeControllerManager *KubeControllerManagerConfig `json:",omitempty"`
-//	KubeScheduler         *KubeSchedulerConfig         `json:",omitempty"`
-//	Docker                *DockerConfig                `json:",omitempty"`
-//	APIServer             *APIServerConfig             `json:",omitempty"`
-//
-//	DNS *DNSConfig `json:",omitempty"`
-//
-//	// NodeConfig can directly access a store of secrets, keys or configuration
-//	// (for example on S3) and then configure based on that
-//	// This supports (limited) dynamic reconfiguration also
-//	SecretStore string `json:",omitempty"`
-//	KeyStore    string `json:",omitempty"`
-//	ConfigStore string `json:",omitempty"`
-//
-//	KubeUser string `json:",omitempty"`
-//
-//	Tags   []string `json:",omitempty"`
-//	Assets []string `json:",omitempty"`
-//
-//	MasterInternalName string `json:",omitempty"`
-//
-//	// The DNS zone to use if configuring a cloud provided DNS zone
-//	DNSZone string `json:",omitempty"`
-//
-//	// Deprecated in favor of KeyStore / SecretStore
-//	Tokens       map[string]string          `json:",omitempty"`
-//	Certificates map[string]*fi.Certificate `json:",omitempty"`
-//	PrivateKeys  map[string]*fi.PrivateKey  `json:",omitempty"`
-//}

upup/pkg/fi/nodeup/loader.go

@@ -124,7 +124,18 @@ func (r *Loader) newTaskHandler(prefix string, builder TaskBuilder) loader.Handl
 		if err != nil {
 			return err
 		}
-		task, err := builder(i.Name, contents, i.Meta)
+		name := i.Name
+		if strings.HasSuffix(name, ".template") {
+			name = strings.TrimSuffix(name, ".template")
+			expanded, err := r.executeTemplate(name, contents)
+			if err != nil {
+				return fmt.Errorf("error executing template %q: %v", i.RelativePath, err)
+			}
+
+			contents = expanded
+		}
+
+		task, err := builder(name, contents, i.Meta)
 		if err != nil {
 			return fmt.Errorf("error building %s for %q: %v", i.Name, i.Path, err)
 		}

upup/pkg/fi/nodeup/local/local_target.go

@@ -3,6 +3,7 @@ package local
 import "k8s.io/kops/upup/pkg/fi"
 
 type LocalTarget struct {
+	CacheDir string
 }
 
 var _ fi.Target = &LocalTarget{}

upup/pkg/fi/nodeup/nodetasks/load_image.go

@@ -0,0 +1,71 @@
+package nodetasks
+
+import (
+	"fmt"
+	"github.com/golang/glog"
+	"k8s.io/kops/upup/pkg/fi"
+	"k8s.io/kops/upup/pkg/fi/hashing"
+	"k8s.io/kops/upup/pkg/fi/nodeup/cloudinit"
+	"k8s.io/kops/upup/pkg/fi/nodeup/local"
+	"k8s.io/kops/upup/pkg/fi/utils"
+	"os/exec"
+	"path"
+	"strings"
+)
+
+// LoadImageTask is responsible for downloading a docker image
+type LoadImageTask struct {
+	Source string
+	Hash   string
+}
+
+var _ fi.Task = &LoadImageTask{}
+
+func (t *LoadImageTask) String() string {
+	return fmt.Sprintf("LoadImageTask: %s", t.Source)
+}
+
+func (e *LoadImageTask) Find(c *fi.Context) (*LoadImageTask, error) {
+	glog.Warningf("LoadImageTask checking if image present not yet implemented")
+	return nil, nil
+}
+
+func (e *LoadImageTask) Run(c *fi.Context) error {
+	return fi.DefaultDeltaRunMethod(e, c)
+}
+
+func (_ *LoadImageTask) CheckChanges(a, e, changes *LoadImageTask) error {
+	return nil
+}
+
+func (_ *LoadImageTask) RenderLocal(t *local.LocalTarget, a, e, changes *LoadImageTask) error {
+	hash, err := hashing.FromString(e.Hash)
+	if err != nil {
+		return err
+	}
+
+	url := e.Source
+
+	localFile := path.Join(t.CacheDir, hash.String()+"_"+utils.SanitizeString(url))
+	_, err = fi.DownloadURL(url, localFile, hash)
+	if err != nil {
+		return err
+	}
+
+	// Load the image into docker
+	args := []string{"docker", "load", "-i", localFile}
+	human := strings.Join(args, " ")
+
+	glog.Infof("running command %s", human)
+	cmd := exec.Command(args[0], args[1:]...)
+	output, err := cmd.CombinedOutput()
+	if err != nil {
+		return fmt.Errorf("error loading docker image with '%s': %v: %s", human, err, string(output))
+	}
+
+	return nil
+}
+
+func (_ *LoadImageTask) RenderCloudInit(t *cloudinit.CloudInitTarget, a, e, changes *LoadImageTask) error {
+	return fmt.Errorf("LoadImageTask::RenderCloudInit not implemented")
+}

upup/pkg/fi/nodeup/nodetasks/service.go

@@ -41,7 +41,7 @@ func (p *Service) GetDependencies(tasks map[string]fi.Task) []fi.Task {
 		// We assume that services depend on basically everything
 		typeName := utils.BuildTypeName(reflect.TypeOf(v))
 		switch typeName {
-		case "*CopyAssetTask", "*File", "*Package", "*Sysctl", "*UpdatePackages", "*User", "*Disk":
+		case "*CopyAssetTask", "*File", "*Package", "*Sysctl", "*UpdatePackages", "*UserTask", "*Disk":
 			deps = append(deps, v)
 		case "*Service":
 		// ignore